do older investors make better investment decisions? · investment decisions of older investors,...
TRANSCRIPT
Electronic copy available at: http://ssrn.com/abstract=767125
DO OLDER INVESTORS MAKE
BETTER INVESTMENT DECISIONS?
George M. Korniotis and Alok Kumar∗
Abstract – This paper examines the investment decisions of older individual investors. We find that older
and experienced investors are more likely to follow “rules of thumb” that reflect greater investment
knowledge. However, older investors are less effective in applying their investment knowledge and
exhibit worse investment skill, especially if they are less educated, earn lower income, and belong to
minority racial/ethnic groups. Overall, the adverse effects of aging dominate the positive effects of
experience. These results indicate that older investors’ portfolio decisions reflect greater knowledge
about investing but investment skill deteriorates with age due to the adverse effects of cognitive aging.
(JEL D14, G11, J14)
I. Introduction
The older population in the United States is growing at a dramatic pace and it is also becoming
more diverse in terms of its racial and ethnic composition.1 Because of this growth in the
proportion of older people, there has been heightened interest in understanding their post-
∗Korniotis: Board of Governors of the Federal Reserve System; Kumar: McCombs School of Business, Uni-
versity of Texas at Austin. We would like to thank two anonymous referees, Warren Bailey, Robert Battalio, Jeff
Bergstrand, Sudheer Chava, George Constantinides, Shane Corwin, Tom Cosimano, Alex Edmans, Joe Egan,
Xavier Gabaix, John Griffin, David Hirshleifer, Roger Huang, Mark Hulbert, Zoran Ivkovich, Jerry Langley,
Sonya Lim, Tim Loughran, Alex Michaelides, Fred Mittelstaedt, David Ng, David Pearce, Jim Poterba, Paul
Schultz, Bob Shiller, John Stiver, Paul Tetlock, Stathis Tompaidis, Mitch Warachka, Mark Watson (the editor),
Frank Yu, Eduardo Zambrano, Margaret Zhu, and seminar participants at Notre Dame, University of Amster-
dam, NHH Bergen, BI Norweigian School of Management, and University of Cyprus for helpful discussions and
valuable comments. In addition, we would like to thank Itamar Simonson for making the investor data available
to us and Brad Barber and Terrance Odean for answering numerous questions about the investor database. We
are responsible for all remaining errors and omissions. This paper previously circulated under the title “Does
Investment Skill Decline due to Cognitive Aging or Improve with Experience?”.
1
Electronic copy available at: http://ssrn.com/abstract=767125
retirement quality of life. The popular media has often raised the concern that older people
would not be able to generate the annual income necessary to sustain the pre-retirement quality
of life. Thus, as the U.S. population ages, it becomes important to understand the investment
decisions of older individual investors because investment income is likely to be a significant
proportion of their post-retirement income, and therefore, one of the main determinants of their
post-retirement quality of life.
In this study, we focus on an important but previously unexplored determinant of the stock
investment decisions of older investors, namely, cognitive aging. We examine whether older
people make better investment choices as they gain more investment knowledge and experience,
or whether their investment skill deteriorates with age due to the adverse effects of cognitive
aging. This is an important issue that has implications for how individual investors should
structure their portfolios over time, the type of investment advice they should seek over their
lifetime, and the potential effects of changes in government policy on investment generated
retirement income. To our knowledge, this is the first study that highlights the potential role
and the importance of cognitive aging in people’s investment decisions.
The extant evidence from cognitive aging and learning research indicates that aging and
learning processes operate jointly. In particular, studies in cognitive aging demonstrate that
both physical and cognitive abilities, especially memory, decline with age (e.g., Horn (1968), Fair
(1994, 2004), Salthouse (2000), Schroeder and Salthouse (2004)). Further, research in learning
suggests that with experience, older investors might accumulate greater investment knowledge
and exhibit greater awareness of the fundamental principles of investing. Their accumulated
investing wisdom could help them make better investment decisions. Investors might also be
less prone to behavioral biases as they grow older and become more experienced (e.g., List
(2003), Feng and Seasholes (2005), Dhar and Zhu (2006), Goetzmann and Kumar (2008)).
Motivated by these earlier findings, we conjecture that, on the one hand, older investors
would accumulate greater knowledge about the fundamental principles of investing. But on the
other hand, their declining cognitive abilities would hinder the effective application of those
2
principles. If the adverse effects of aging dominate the positive effects of experience, older
investors’ portfolios would under-perform common performance benchmarks. Using the end-of-
month portfolio holdings and trades of a sample of individual investors at a large U.S. brokerage
house, we empirically test this dual pronged conjecture.
The empirical analysis focuses on the relative influences of age and investment experience on
investors’ portfolio decisions and performance. We estimate “rules of thumb” and “skill” regres-
sions, where the dependent variable is either a measure that reflects the outcome of following
an investment rule of thumb or a measure of investment skill. The key explanatory variables
are age and investment experience. We use age to capture the adverse effects of cognitive aging
and use experience (the number of days between the account opening date and December 31,
1996) to capture the positive effects of experience.2 Without the experience measure, age would
capture two confounding effects, one related to experience and the other related to cognitive
aging. By including age and experience variables simultaneously, we attempt to separate the
positive effects of experience from the negative effects of cognitive aging.
Our empirical evidence strongly supports our main conjecture. Consistent with the theo-
retical predictions of life-cycle and learning models, we find that older and more experienced
investors hold less risky portfolios, exhibit stronger preference for diversification, trade less fre-
quently, exhibit greater propensity for year-end tax-loss selling, and exhibit weaker behavioral
biases. And consistent with the psychological evidence, we find that older investors exhibit
worse stock selection ability and poor diversification skill. The age-skill relation has an inverted
U-shape and, furthermore, the skill deteriorates sharply around the age of 70. These results
suggest that older investors exhibit a greater propensity to use common investment rules of
thumb but they appear less skillful in successfully implementing those rules.3
To gather additional support for our main hypothesis, we seek guidance from the psychol-
ogy literature that examines the conditional effects of cognitive aging. The evidence from this
literature indicates that people who are more educated, more resourceful, and undertake intel-
lectually stimulating jobs are able to better compensate for their declining cognitive abilities
3
(Arbuckle, Gold, and Andres (1986), Baltes and Lang (1997), Cagney and Lauderdale (2002)).
The evidence also suggests that the age-related decline in cognitive abilities is steeper for African
Americans and Hispanic minority groups (e.g., Avolio and Waldman (1986), Black (2004)).
Motivated by these findings in cognitive psychology, we use age-income, age-education, age-
race, and age-ethnicity interaction terms as additional proxies for cognitive aging and examine
the conditional deterioration in cognitive abilities with age. Consistent with our hypothesis, we
find that the adverse effects of aging are stronger among older investors who are also relatively
less educated, earn lower income, and belong to the Hispanic ethnic group.
Because we cannot measure the adverse effects of cognitive aging directly, our results are
open to alternative interpretations. To further ensure that the variables we use to capture
cognitive aging are appropriate, we estimate a model of cognitive aging using a representative
European household-level data set, which includes direct measures of cognitive abilities. The
model estimates indicate that cognitive abilities increase with education, wealth, and income
but decline with age. There is a sharp decline after the age 70 and, moreover, the cognitive
decline is steeper among older individuals who are also less educated and have lower income.
These results indicate that demographic variables such as age, income, wealth, education, and
their interactions are likely to capture the adverse effects of cognitive aging reasonably well.
Studies like ours that examine the effects of age also face the classic age-cohort-period iden-
tification problem (e.g., Deaton (1997), Ameriks and Zeldes (2004)). A potential alternative
interpretation of our evidence on cognitive aging is that it reflects birth cohort effects. While
plausible, there are several reasons why cohort effects are unlikely to explain our main findings.
First, it is difficult to conceive a hypothesis based on cohort effects that predicts the observed
opposite effects of age in our rules of thumb and skill regressions. Put differently, cohort effects
cannot easily explain why older investors would exhibit a greater propensity to follow common
investment principles but exhibit a lower ability to apply them effectively. Second, cohort-based
explanations for the observed sudden drop in performance at older age are unlikely to be con-
vincing. Third, cohort effects cannot successfully account for the inverted U-shaped relation
4
between age and investment skill. Last, the performance differential between younger and older
investors increase with portfolio size, which we interpret as a proxy for task difficulty. There is
no obvious cohort-based explanation for this result. In contrast, all these findings are strongly
consistent with the cognitive aging hypothesis and reflect the natural outcome of the joint aging
and learning processes.
To further account for the effects of birth cohorts, we perform an individual-level comparison
and use the differencing method proposed in McKenzie (2006). For each investor, we compute
the change in performance between the first and the second halves of the sample period and
examine whether the performance differential varies with age. By differencing, we eliminate
the common effects associated with cohorts. We find that the performance differential measure
exhibits an inverted U-shaped pattern and, similar to the age-skill relation, there is a sharp
decline at very old age. This evidence is consistent with the cognitive aging hypothesis and does
not suffer from potential contamination from cohort effects.
Examining the economic costs of aging, we find that, on average, investors with stronger
aging effects earn about 3% lower risk-adjusted annual returns, and the performance differential
is over 5% among older investors with large portfolios. These performance differentials remain
economically significant even when we account transaction costs and use net returns to measure
performance. We conduct several robustness tests and show that our results are not sensitive
to the relatively short sample size, exceptional performance of certain styles and industries, the
choice of skill measures, potential error in skill measurement, our inability to observe investors’
full portfolios, choice of the risk adjustment methodology, specific market conditions, investors’
lack of interest in relatively small portfolios, and the geographical concentration of our sample
investors.
At a first glance, these empirical findings might appear puzzling, and perhaps surprising,
because the finance literature on portfolio choice mainly attributes increasing risk aversion and
the positive effects of experience to the aging process. The adverse effects of cognitive aging are
typically ignored. Within this traditional portfolio choice paradigm, it is very difficult to conceive
5
a hypothesis that posits a positive impact of experience and the negative impacts of age, age-
income, age-education, and age-race/ethnicity interaction terms on investment skill. However,
when interpreted in the appropriate context of the extant psychological evidence on cognitive
aging, our findings appear intuitive and economically meaningful because they represent the
natural outcome of the joint aging and learning processes.
The rest of the paper is organized as follows. In the next section, we develop the main
testable hypotheses. We describe the data in Section III and, in Section IV, we test our first
hypothesis that focuses on the positive effects of aging and experience. In Section V, we test our
unconditional and conditional hypotheses, which posit that stock investment skill deteriorates
with age but improves with experience. In Section VI, we estimate the economic costs of aging.
In Section VII, we conduct robustness checks and attempt to rule out alternative interpretations
of our key findings. We conclude in Section VIII with a brief summary of the paper and potential
implications of our results.
II. Hypothesis Development and Related Research
We develop our testable hypotheses by synthesizing the empirical evidence from the psychological
research on aging, the literature on learning, and life-cycle models of investing. The extant
psychological evidence indicates that the decline in cognitive abilities is a normal consequence
of biological aging and this phenomenon is observed across different countries and cultures (Craik
and Salthouse (1992)).
Both physical and cognitive abilities, especially memory, decline with age (e.g., Horn (1968),
Fair (1994, 2004), Salthouse (2000), Schroeder and Salthouse (2004)), where the decline begins
at a relatively young age of 30 (Grady and Craik (2000)). Weakening memory slows down the
information processing ability of individuals and leads to a decline in older people’s ability to
perceive conditional probabilities (Spaniol and Bayen (2005)). In addition, due to a decline
in attentional ability, older people get distracted easily and are unable to distinguish between
6
relevant and irrelevant information. Overall, the evidence suggests that older people would
react to new information inappropriately because they are typically slower and less effective at
processing and integrating new information.
The psychological evidence also indicates that people are likely to experience a decline in
the level of their general intelligence as they grow older. The aging process influences general
intelligence through two distinct channels. First, general intelligence declines with age due to the
adverse effects of aging on memory and attention (e.g., Lindenberger and Baltes (1994), Baltes
and Lindenberger (1997)). Second, the sensory (vision and hearing) functioning worsens with
age and is associated with lower levels of intelligence. The decline in intelligence is much steeper
after the age of 70 (Lindenberger and Baltes (1997)), while these adverse effects are attenuated
in people’s area of expertise due to frequent practicing (Masunaga and Horn (2001)).
In addition to biological and psychological factors, socioeconomic and demographic factors
such as education, income, wealth, race, ethnicity, and gender can exacerbate the adverse effects
of cognitive aging. For example, people who are more educated, more resourceful (i.e., have
higher income and are wealthier), and undertake intellectually stimulating jobs experience a
slower decline in cognitive abilities because they are able to actively compensate for the adverse
effects of aging (Arbuckle, Gold, and Andres (1986), Baltes and Lang (1997), Cagney and
Lauderdale (2002)). In contrast, the age-related decline in cognitive abilities is steeper among
older women (Shanan and Sagiv (1982)) as well as older African Americans and Hispanics
(Avolio and Waldman (1986), Black (2004)).
While old age is likely to have an adverse effect on people’s ability to make effective in-
vestment decisions, older investors are likely to have greater investment experience and greater
awareness of the fundamental principles of investing. Their accumulated investing wisdom could
help them make more efficient investment decisions. Theoretical models of portfolio choice (e.g.,
Bakshi and Chen (1994), Campbell and Viceira (2002), Cocco, Gomes, and Maenhout (2005),
Gomes and Michaelides (2005)) also posit that the riskiness of investor portfolios would decline
with age due to decreasing investment horizon and increasing risk aversion.4
7
In addition to these channels, investors are likely to learn through the process of trading,
and they might be less prone to behavioral biases as they grow older and become more experi-
enced. The extant empirical evidence from the individual investor literature indicates that older
investors exhibit a weaker disposition effect (Dhar and Zhu (2006)), hold less concentrated port-
folios (Goetzmann and Kumar (2008)), and exhibit lower degree of over-confidence (Barber and
Odean (2001)). Furthermore, these behavioral biases decline as investors learn and gain more
experience (e.g., List (2003), Feng and Seasholes (2005), Goetzmann and Kumar (2008)). Older
investors are also less prone to gambling type activities in the stock market (Kumar (2009)).
Overall, the consensus that emerges from cognitive aging and learning research suggests that,
on the one hand, older investors are likely to make relatively more conservative choices and might
possess relatively greater knowledge about the fundamental principles of investing. But on the
other hand, effective application of those principles requires efficient processing of information,
which they might lack.5 Given the opposite predictions of aging and learning research, our paper
investigates empirically whether declining cognitive abilities or increasing investment experience
has a stronger influence on investors’ stock investment decisions.
We test three hypotheses. First, based on the evidence from learning studies, we posit that:
H1: Investment knowledge increases with both age and experience.
Next, based on the extant psychological evidence, we develop our unconditional hypothesis,
which posits that:
H2: Investment skill increases with experience due to the positive effects of learning
but declines with age due to the adverse effects of cognitive aging.
While support for the unconditional hypothesis would provide evidence consistent with the
predictions of cognitive aging, in the absence of a direct measure of cognitive aging, the following
conditional hypothesis would strongly reinforce that evidence:
H2cond: The decline in investment skill is steeper among older investors who are
less educated, earn lower income, and belong to a racial/ethnic minority group.
8
In the following sections, we test these hypotheses using data from multiple sources.
III. Data and Sample Selection
The main data set for this study consists of all trades and end-of-month portfolio positions of
investors at a major U.S. discount brokerage house for the 1991 to 1996 time period. There are
77,995 households in the retail database who hold common stocks and trade a variety of other
securities including mutual funds, options, American depository receipts (ADRs), etc. In this
study, we focus on the investment behavior of 62,387 investors who have traded common stocks.
For a subset of households, demographic information such as age, income, wealth, occupation,
marital status, gender, etc. is available. The demographic measures such as age, income, marital
status, family size, etc. were compiled by Infobase Inc. a few months after the end of the sample
period (June 1997). The total wealth is reported at the account opening date.6
An average investor holds a four-stock portfolio (median is three) with an average size of
$35,629 (median is $13,869). The average aggregate value of investor portfolios in our sample
is about 2.18 billion and our sample investors executed about 1.9 million common stock trades
during the six-year sample period. Investors’ stock holdings and trades encompass about 90% of
stocks (9,011 stocks) from the Center for Research on Security Prices (CRSP) universe. Further
details on the investor database are available in Barber and Odean (2000).
Table I, Panel A reports summary statistics for five groups of investors grouped as age
cohorts. Because our primary focus is on the investment behavior of older investors, in Panel
B, we also provide summary statistics for five groups of older investors (age ≥ 60) grouped
as age cohorts. A typical investor in our sample has held a brokerage account for about ten
years, and as expected, investment experience increases with age. More importantly, we find
that consistent with the prior evidence (e.g., Poterba (2001)), the mean portfolio size increases
monotonically with age and there is no evidence that older investors reduce their exposure to
equity as their investment horizon decreases. In fact, older investors have greater proportional
9
investment in the stock market, both when measured as a proportion of their total wealth and
their annual income.
The cross-sectional variations in wealth and income in our sample also match well with
corresponding cross-sectional variations in the more representative Survey of Consumer Finances
(SCF) data. For instance, consistent with the evidence in Poterba (2001), we find that the wealth
level peaks within the age range of 65-69. Additionally, we find that the annual income peaks
within the age range of 47-52, which is also consistent with the predictions of the life-cycle
models. These comparisons with the SCF data suggest that our sample of older retail investors
is reasonably representative of the older households in the U.S.7
To enrich our analysis, we complement the individual investor data with demographic data
from the 1990 U.S. Census. We use the Census data to identify the racial/ethnic profile and
the educational background of investors. To identify the racial/ethnic composition of investors,
for each zip code, we compute the proportion of individuals in the following four racial/ethnic
groups: (i) Caucasian, (ii) African American, (iii) Hispanic, and (iv) Others. Using the zip code
of each investor, we assign her the appropriate zip code-level racial/ethnic profile. Similarly,
we use the Census data to infer the education level of an investor. Investors who live in more
educated zip codes are assumed to be more educated, where the proportion of the zip code
population that holds a bachelor’s or higher degree is used to identify the educational status of
that zip code.
Several other data sets are used in this study. We use a representative household-level data
set from the 2005 wave of the Survey of Health, Aging, and Retirement in Europe (SHARE) to
estimate a model of cognitive abilities. We use data from the 2005 Dutch DNB Household Survey
for some of our robustness checks. For each stock in the sample, we obtain the quarterly cash
dividend payments, monthly prices, returns, and market capitalization data from the Center
for Research on Security Prices (CRSP) and quarterly book value of common equity data from
COMPUSTAT. We obtain the monthly time-series of the three Fama-French factors and the
momentum factor from Professor Kenneth French’s data library. Last, we obtain characteristic-
10
based performance benchmarks from Professor Russ Wermer’s web site.8
IV. Age, Experience, and Investment Knowledge
In this section, we gather support for our first main hypothesis (H1). Specifically, we examine
whether the knowledge and experience accumulated over time are reflected in investors’ portfolio
holdings and trading decisions.
A. Age and Equity Portfolio Risk
In our first formal test, we investigate whether older investors tilt their portfolios toward rel-
atively less risky stocks. We estimate panel regression models to examine the characteristics
of age based group portfolios in a multivariate setting. In these regressions, the excess weight
assigned to a stock in the aggregate group portfolio is the dependent variable and the mean
return, idiosyncratic volatility, skewness, kurtosis, and the price of the stock are used as the
primary independent variables.9 The group portfolio is constructed by combining the portfolios
of all investors who belong to the group. The return moments are calculated using the past 60
months of data and the idiosyncratic volatility measure is the variance of the residual obtained
by fitting a four-factor model to the monthly stock returns series. Additionally, we consider the
following control variables to characterize investors’ stock preferences: (i) market beta, which is
also estimated using past 60 months of data, (ii) firm size, (iii) book-to-market ratio, (iv) short-
term momentum (past one-month stock return), (v) longer-term momentum (past twelve-month
stock return), (vi) an S&P500 dummy that is set to one if the stock belongs to the S&P500
index, (vii) monthly volume turnover, and (viii) annual dividend yield.
We account for potential auto-correlation and cross-correlation in errors using the non-
parametric approach of Driscoll and Kraay (1998) and obtain the corrected standard errors
for our estimates. This methodology provides a unified approach for simultaneously correcting
the standard errors for cross-sectional correlation, serial correlation, and cross serial correlation
11
in a panel setup. To ensure that extreme values are not affecting our results, we winsorize all
variables at their 0.5 and 99.5 percentile levels. Further, to facilitate comparisons among the
coefficient estimates, the independent variables are standardized so that each variable has a
mean of zero and a standard deviation of one.
The panel regression estimates are presented in Table II. We estimate the panel regression
model for three age-based categories: (i) younger age group containing investors in the 20-38
age range, (ii) older age group consisting of investors in the 60-94 age range, and finally, (iii)
within the older investor group, the oldest age group consisting of investors in the 75-94 age
range. Additionally, we estimate two panel regressions to examine the differences in the stock
preferences of groups (i) and (ii), and (i) and (iii).
Focusing on the differential regression estimates (see columns (4) and (5)), we find that older
investors favor relatively less risky stocks. Specifically, older investors’ preferences for stocks with
higher idiosyncratic volatility, higher market beta, lower market capitalization, lower prices are
weaker than those of younger investors. In addition, older investors exhibit weaker preference
for stocks with higher skewness, which indicates they are less likely to chase extreme positive
returns. Our estimates also indicate that investors’ preferences for less risky stocks increase with
age. The differences in the stock preferences of younger and older investors, as reflected by the
magnitudes of the coefficient estimates, are stronger when we consider the 75-94 age group to
identify older investors (see column (5)).10
Collectively, the panel regression estimates indicate that the average risk exposures of in-
vestors’ stock portfolios decrease with age. Consistent with our first hypothesis (H1), the ev-
idence indicates that experienced and prudent investors reduce the riskiness of their portfolios
as they grow older.
B. Do Investors Accumulate More Knowledge As They Grow Older?
To examine whether older investors possess greater knowledge about investing, we concentrate
on several important dimensions of portfolio decisions that reflect common investment “rules
12
of thumb”. The set of decisions that we consider is guided by the availability of data. First,
we examine whether older investors are more likely to recognize the potential benefits of diver-
sification. Next, we examine whether older investors trade less frequently because they realize
that they would not be able to improve performance through active trading. Last, we examine
whether older investors are more likely to engage in year-end tax-loss selling, since it requires
financial awareness but does not necessarily require skill.11
We estimate several rule of thumb cross-sectional regressions, where the dependent variable
is a measure of investment decision that reflects a specific investing rule of thumb. The measures
are obtained for each investor using their decisions during the entire sample period. For each
investment decision, we estimate both unconditional and conditional regression models. In the
unconditional regressions, the independent variables are only age and investment experience.
Age corresponds to the head of the household and Investment Experience is the number of days
between the account opening date and December 31, 1996.
In the conditional regressions, the following demographic variables and portfolio charac-
teristics are employed as control variables: Income represents the annual household income;
Education represents the proportion of people in investor’s zip code that has attained a bach-
elor’s or higher educational degree; Male Dummy is set to one if the head of the household
is male; Retired dummy is set to one if the head of the household is retired; Portfolio Size is
the average market capitalization of the household portfolio; Portfolio Turnover is the average
of monthly buy and sell turnovers; and Portfolio Dividend Yield is the average dividend yield
of the household portfolio. The RMRF (market factor), SMB (small-minus-big factor), HML
(high-minus-low factor), and UMD (up-minus-down or momentum factor) exposures are the
factor loadings of an investor’s portfolio and characterize the riskiness of the portfolio. The
factor loadings are obtained by fitting a four-factor time-series model to the monthly portfolio
return series of each investor for the period in which the investor is active. Last, Mutual Fund
Holdings is the proportion of the equity portfolio that is allocated to mutual funds.
The rule of thumb cross-sectional regression estimates are presented in Table III, where the
13
t-statistics are computed using robust and zip code-level clustered standard errors. As before,
we winsorize all variables at their 0.5 and 99.5 percentile levels and the independent variables
are standardized. In the first two regressions, we use the number of stocks in the portfolio as the
dependent variable to examine whether older investors are more aware of the potential benefits
of diversification (columns (1) and (2)). Our intent here is not to show that older investors hold
more diversified stock portfolios. Given that we cannot observe the entire portfolios of investors,
the number of stocks is likely to be a very noisy measure of diversification. Nonetheless, the
number of stocks in an investor’s equity portfolio is likely to indicate whether or not an investor
exhibits a preference for diversification and at least attempts to diversify.12
Our results from the unconditional model indicate that older and more experienced investors
hold portfolios containing a greater number of stocks. The coefficient estimates are significantly
positive for both Age and Investment Experience, even in the presence of various control vari-
ables. The coefficient estimates for the control variables are also broadly consistent with our
expectations. For instance, the positive coefficient estimate for Mutual Fund Holdings indicates
that investors who exhibit stronger preference for diversification in their stock portfolios also
hold more mutual funds. Overall, consistent with our first hypothesis, the evidence indicates
that the portfolio choices of older and more experienced investors are more likely to conform to
one of the key principles of investing, namely, portfolio diversification.
Next, we examine whether older investors follow one of the key recommendations of the effi-
cient market hypothesis, which posits that investors cannot improve performance through active
trading. In the next two regression specifications (columns (3) and (4)), we use the monthly
portfolio turnover as the dependent variable.13 Again, consistent with our first hypothesis, we
find that both older and more experienced investors exhibit lower turnover rates. The coefficient
estimates are significantly negative for both Age and Investment Experience, even in the presence
of various control variables. The negative coefficient estimate for Age and the positive coefficient
estimate for the Male Dummy are consistent with the evidence in Barber and Odean (2001),
who find that older (male) investors trade relatively less (more) frequently. These estimates
14
indicate that the trading behavior of older investors are more likely to conform to another key
principle of investing, namely, less frequent trading.
While these cross-sectional regression estimates are consistent with our first hypothesis, one
might conjecture that a stronger preference for diversification and lower portfolio turnover do not
necessarily imply greater investment knowledge but merely reflect “passive” investing by older
investors. For example, an older investor might hold more stocks in her portfolio not because
she has a preference for diversification, but because she has accumulated a large number stocks
during a long investing period. To address this potential concern, we use a knowledge measure
that is more likely to capture investor’s knowledge about investing. Specifically, we examine
whether older investors exhibit a greater propensity to engage in year-end tax-loss selling.
We estimate a regression model in which the proportion of “losers” (stock investments in
which an investor suffers a loss) sold in the month of December is used as a dependent variable.
The regression estimates indicate that both older and more experienced investors are more will-
ing to sell their losers in December (see column (5)). The coefficient estimates are significantly
positive for both Age (estimate = 0.016, t-stat = 5.696) and Investment Experience (estimate
= 0.006, t-stat = 2.311), even in the presence of various control variables. Again, we find that
most control variables have the expected signs. For instance, investors who hold larger portfolios
and trade more often are likely to sell more losers in December. This evidence indicates that
the trading behavior of older investors reflect yet another investing rule of thumb: “Sell your
losers in December”.
Given that the coefficient estimates of Age and Investment Experience have similar signs in
all cross-sectional regression specifications, one might suspect that the two variables capture
identical aspects of investors’ investment decisions. But we find that age and investment expe-
rience measures are weakly correlated (correlation = 0.142). Furthermore, when we estimate a
cross-sectional regression with portfolio dividend yield as the dependent variable, we find that
the coefficient estimate for Age is positive but the coefficient estimate for Investment Experience
is negative (see column (6)). The positive coefficient estimate for Age is consistent with the
15
evidence in Graham and Kumar (2006), who show that older investors are likely to prefer high
yield stocks for consumption reasons. However, the negative estimate for Investment Experience
is new and it indicates that, all else equal, more experienced investors prefer lower yield stocks.
In the current context, more importantly, the estimates indicate that the age and experience
variables capture two distinct aspects of investors’ portfolio choices and trading decisions.
C. Other “Rules of Thumb”
While it is very difficult to perform a comprehensive analysis of rules of thumb recommended
by investment advisors, we study whether older and experienced investors exhibit a greater
propensity to follow other commonly prescribed rules of thumb. The results from these addi-
tional empirical tests are summarized in Table III, Panel B. First, we examine whether older
investors exhibit a higher propensity to spread their positions across multiple asset classes. We
also investigate whether older investors are more likely to diversify using low expense mutual
funds and foreign stocks.
Consistent with the evidence on diversification preference, the estimates from test (1) show
that both older and experienced investors invest in a greater number of asset classes. We also find
that both older and experienced investors hold mutual funds with lower expense ratios and hold a
larger proportion of foreign stocks in their equity portfolios (see tests (2) and (3)).14 For example,
the expense ratio differential between a 30 year old and a 65 year old investor is 0.078%, which
is about 9.10% higher than the mean expense ratio of sample investors (= 0.857%). Similarly,
older investors allocate 2.02% more weight to foreign stocks than younger investors. This is more
than 50% higher than the sample mean foreign stock weight of 3.87%. These results indicate
that older investors are more likely to follow two other common investment rules of thumb:
“Invest in well-diversified, low expense mutual funds” and “Diversify internationally.”15
Next, in tests (4) and (5), we examine whether both older and experienced investors exhibit
weaker behavioral biases such as the disposition effect and local or familiarity bias. We find that
consistent with the greater investment knowledge hypothesis, older and experienced investors
16
are less likely to hold on to their losers (i.e., they exhibit a weaker disposition effect) and exhibit
a relatively lower propensity to hold local stocks (i.e., exhibit a weaker local bias).16
Last, we examine whether older investors shy away from relatively sophisticated trading
strategies such as margin trading, short selling, and option trading (see tests (6) to (8)). We
find that older investors are less likely to hold a short position or trade options. In contrast,
experienced investors are more likely to hold short positions and they also exhibit a greater
propensity to trade options. Investors with greater experience are also more likely to have a
margin trading account. These findings suggest that experience might be a proxy for financial
sophistication and could be positively correlated with investment skill.
D. Evidence From the Dutch DNB Data
To further examine the robustness of our rules of thumb regression results, we consider data
from another country. We use the Dutch DNB data and examine whether older and experienced
investors exhibit stronger diversification motives. The DNB data contain information about the
aggregate holdings in stocks, bonds, and mutual funds but do not provide information about the
types of assets investors hold or trade. We also have information about the respondent’s age, in-
come, education, gender, self-reported knowledge of financial markets (our proxy for investment
experience), self-reported cognitive ability measures, and other demographic variables.
Using the mutual fund participation dummy as a proxy for the intent to diversify, we estimate
a logit model, where the mutual fund participation dummy is the dependent variable and the
various demographic variables including age, experience, and cognitive ability are the indepen-
dent variables.17 The results are reported in Table III, Panel C. Consistent with the results from
the U.S. brokerage data, we find that older and more knowledgeable Dutch investors exhibit a
higher probability of holding mutual funds (see test (1)). The results are similar when we use an
alternative measure to capture investors’ diversification motives: an “All Asset Classes” dummy
that is set to one if an investor holds stocks, bonds, and mutual funds. We find that both Age
and Experience have significantly positive coefficient estimates (see test (2)).
17
When we include the cognitive ability index as an additional dependent variable, in unt-
abulated results, we find that higher levels of cognitive ability increase the mutual fund par-
ticipation probability (estimate = 0.321, z-stat = 2.225). High cognitive ability investors also
exhibit a greater propensity to invest in all asset classes, although the coefficient estimate is
only marginally significant (estimate = 0.577, z-stat = 1.671). These results indicate that like
cognitive abilities, age and experience are positively correlated with investment knowledge.
Overall, our rule of thumb cross-sectional regression results indicate that older investors
make more conservative stock investment choices and their choices reflect greater knowledge
about investing. This evidence supports our first hypothesis (H1), which posits that investment
knowledge increases with both age and experience. Furthermore, the evidence indicates that
investors with greater experience appear more sophisticated and could be skilled.
V. Cognitive Aging and Investment Skill
While older investors, especially those who are more experienced, exhibit a greater propensity
to follow common investment rules of thumb, how effectively can they apply those principles? In
this section, we examine the relation between age, investment experience, and investment skill
and gather support for our second set of hypotheses (H2 and H2cond), which posit opposite
effects of age and experience on investment skill.
A. Graphical Evidence: Univariate Results
Figure 1 shows the univariate relation between age and investment skill, as captured by the
Daniel, Grinblatt, Titman, and Wermers (1997) characteristic-adjusted performance measure.
Two features of the plot are noteworthy. First, the investment performance exhibits an inverted
U-shape with a peak at around 42 years. The hump shape reflects the combined effects of
experience and aging. This evidence is consistent with the findings in Agarwal, Driscoll, Gabaix,
and Laibson (2007), who find a similar pattern in the borrowing rates of households in various
18
credit markets. Second, there is an abrupt and significant drop in investment performance
around the age of 70. This nonlinear effect supports our cognitive aging hypothesis and is
consistent with the evidence from studies in psychology that document a steeper cognitive
decline after the age of 70. Overall, the graphical evidence reveals that the interactions between
the aging and learning processes determine the risk-adjusted performance of investor portfolios.
B. Opposite Effects of Age and Experience
We attempt to uncover these interactions by estimating “skill” regressions. In these cross-
sectional regressions, a measure of investment skill is employed as the dependent variable. We
focus on two specific investment skill measures: the “diversification skill” and the stock selec-
tion ability. Our conjecture is that although older investors hold portfolios with larger number
of stocks, they might not possess “diversification skill” because the ability to perceive corre-
lations accurately would decline with age.18 Furthermore, investors’ stock selection skill could
decline with age because the adverse effects of cognitive aging would influence people’s ability
to efficiently process new information. In contrast, both diversification skill and stock selection
abilities would improve with investment experience.
We use the monthly Sharpe ratio to measure diversification skill and the monthly portfolio
alpha to measure stock selection skill. For each investor, the realized portfolio return series
during the full sample period is used to compute the two performance measures. The Sharpe
ratio measure is defined as SRp = (µp−rf)/σp, where µp is the average realized portfolio return,
σp is the standard deviation of the portfolio return series, and rf is the risk-free rate of return.
The alpha measure is the intercept obtained by fitting a four-factor time-series model to the
portfolio return series. The factor model includes the market (RMRF), size (SMB), value (HML),
and momentum (UMD) factors. We also employ the mean characteristic-adjusted monthly
portfolio return measure computed using the Daniel, Grinblatt, Titman, and Wermers (1997)
methodology as an alternative measure of investment skill.19 Both investment skill measures
capture the ability of investors to generate excess returns from their portfolio decisions, after
19
accounting for the known differences in their stock preferences (see Table II).
In the first set of skill regressions, we test our unconditional hypothesis. The only inde-
pendent variables in these regression specifications are Age and Investment Experience. The
cross-sectional regression estimates for the unconditional models are presented in Table IV,
Panel A (columns (1), (3), and (5)). In all our regression specifications, we use robust, clustered
standard errors to account for potential cross-sectional dependence in performance within zip
codes.
In the Sharpe ratio regression (column (1)), Age has a negative but statistically insignificant
coefficient estimate (estimate = −0.001, t-stat = −0.935), while Investment Experience has
a significantly positive coefficient estimate (estimate = 0.011, t-stat = 11.644).20 In the alpha
regression (column (3)), Age has a significantly negative coefficient estimate (estimate = −0.042,
t-stat = −5.047) and Investment Experience has a positive but statistically insignificant estimate
(estimate = 0.010, t-stat = 1.184). Last, in the characteristic-adjusted return regression (column
(5)), Age has a significantly negative coefficient estimate (estimate = −0.044, t-stat = −7.293)
while Investment Experience has a significantly positive coefficient estimate (estimate = 0.037, t-
stat = 6.053). These coefficient estimates are broadly consistent with our second main hypothesis
and indicate that Age has opposite effects in rule of thumb and skill regressions.
The coefficient estimates in the skill regressions are also significant in economic terms. For
instance, the estimate for Age in the alpha regression indicates that, holding experience fixed,
a one standard deviation shift in the age of an investor would correspond to an annual, risk-
adjusted performance decline of 0.042 × 12 = 0.504%. The mean age of investors in our sample
is 50 and the standard deviation is about 12. Thus, when an investor aged 30 becomes older
and crosses the retirement age of 65 (a three standard deviation change in age), she is likely to
suffer an annual performance decline of 1.512% on a risk-adjusted basis.
Collectively, the graphical evidence and the results from our unconditional tests indicate
that, investment skill varies inversely with age and positively with investment experience. This
evidence is consistent with our unconditional hypothesis (H2), although the strengths of age-skill
20
and experience-skill relations are not very strong.21
C. Model of Cognitive Abilities and Skill Regression Specification
We consider additional proxies for cognitive aging to gather stronger support for the second
hypothesis. Our choice is motivated by previous studies in cognitive psychology, which demon-
strate that individuals who are more educated, more resourceful, and undertake intellectually
stimulating jobs are able to better compensate for their declining cognitive abilities (Arbuckle,
Gold, and Andres (1986), Baltes and Lang (1997), Cagney and Lauderdale (2002)). Analogous
to the psychological evidence, we expect that the adverse effects of cognitive aging would be
weaker among wealthier, higher income, and more educated investors. Older investors who are
more educated and more resourceful (i.e., have higher income levels and are wealthier) might
be able to better compensate for their declining information processing abilities.
To ensure that the additional cognitive aging proxies employed in the skill regression specifi-
cation are appropriate, we estimate a model of cognitive abilities using a representative European
household data set and show that our cognitive aging proxies are strongly correlated with direct
measures of cognitive abilities. We choose the European data set because it contains three direct
measures of cognitive abilities: (i) verbal ability, (ii) quantitative ability, and (iii) memory. The
three cognitive measures are positively correlated but the maximum correlation is below 0.50.
Using these measures, we obtain a composite (equal-weighted) measure of cognitive abilities.
The European household data also contain demographic variables that are available in our in-
dividual investor data set. We consider several regression specifications, where one of direct
measures of cognitive abilities is the dependent variable and the main determinants of cognitive
abilities identified in the psychology literature are the independent variables.
The cognitive abilities regression estimates are reported in Table V. Consistent with the
evidence from psychology, we find that cognitive abilities decline with age and are positively
associated with education, wealth, and income. Furthermore, the decline is steeper among
individuals who are considerably older (age > 70), are less educated, and have lower income. It
is interesting that two facets of cognitive abilities that are more relevant for investment decisions,
21
namely, the quantitative ability and memory, exhibit a sharper decline with age. The decline
in verbal ability that might be less relevant for investment decisions is relatively less severe.
Collectively, the cognitive abilities model estimates indicate that demographic variables such as
age, income, wealth, education, and their interactions are likely to capture the adverse effects
of cognitive aging reasonably well.
Motivated by this evidence from our cognitive abilities model, we consider an extended skill
regression model with multiple cognitive aging proxies and attempt to capture the influence
of cognitive aging on investment decisions more accurately. Specifically, we interact age with
income and education, where both the Age×Low Income and Age×Low Education interaction
terms are defined after standardizing the age variable. We also consider an Over 70 Dummy to
capture the sudden drop in investment performance identified in Figure 1.
The evidence from research in cognitive psychology (e.g., Avolio and Waldman (1986), Black
(2004)) suggests that the age-related decline in cognitive abilities is steeper among ethnic/racial
minorities (African Americans and Hispanics). In light of this evidence, we use two additional
interaction terms, one for Hispanics and another for African-Americans. The Hispanic variable
is set to one for investors who live in zip codes where the ratio of the populations of Hispanics
to Whites is in the upper quintile. The African American variable is defined in an analogous
manner. We interact both race/ethnicity variables with Age.
Similar to the unconditional skill regressions, we consider several independent variables to
control for the known determinants of portfolio performance and investment skill. We include a
Male Dummy as a control variable because previous studies have shown that gender influences
net investment performance (Barber and Odean (2001)). The Retired Dummy allows us to
control for the significant lifestyle changes at the time of retirement, which could alter investment
behavior. We employ several portfolio characteristics as control variables because investors’ risk
preferences are likely to vary with age. This set includes portfolio size, portfolio dividend yield,
and the four factor exposures (RMRF , SMB, HML, and UMD coefficient estimates) of the
investor portfolio. Last, the Portfolio Turnover measure at least partially accounts for the effects
22
of transaction costs on portfolio performance.
D. Intent to Diversify versus Actual Diversification Skill
The diversification skill regression estimates for the conditional model is reported in Table
IV, Panel A, Column (2). Consistent with our second hypothesis, we find that the age-skill
and experience-skill relations become stronger and appear more transparent in the conditional
model. In the Sharpe ratio regression, both Age and Investment Experience have significant
coefficient estimates, where the loading on Age is negative (estimate = −0.014, t-stat = −3.355)
and the loading on Investment Experience is positive (estimate = 0.010, t-stat = 9.450). We
interpret the negative coefficient estimate on Age as the adverse effect of cognitive aging, while
the positive coefficient estimate on Investment Experience indicates that greater experience leads
to greater diversification skill.
Our results from the Sharpe ratio regression indicate that the coefficient estimate of Income
is positive but statistically insignificant. However, the Age×Low Income interaction term has
a negative coefficient estimate, which indicates that the adverse effects of aging are stronger
among older investors with lower income. In addition, we find that Education has a significantly
positive coefficient estimate (estimate = 0.002, t-stat = 1.882) and the Age×Low Education
interaction term has a marginally negative coefficient estimate. These results indicate that
while education and investment experience leads to more effective diversification, as investors
grow older, their ability to diversify effectively decreases.
Our results are similar when we use the average correlation among stocks in the portfolio
as an alternative measure of diversification skill. If investors diversify effectively, all else equal,
they would hold stocks that have lower correlations among them. When the average stock
correlation is the dependent variable in the skill regression, in untabulated results, we find
that Age has a significantly positive coefficient estimate (estimate = 0.006, t-stat = 3.032),
Investment Experience has a significantly negative coefficient estimate (estimate = −0.004, t-
stat = −3.903), and other coefficient estimates are very similar to the Sharpe ratio regression
23
estimates. This evidence indicates that although both older and experienced investors attempt
to diversify, only experienced investors demonstrate an ability to achieve effective diversification
and possess diversification skill.
We also obtain very similar results when we evaluate investors’ mutual fund choices. The
rules of thumb regressions in Section IV.C indicate that older investors are more likely to follow
the rule: “Invest in well-diversified, low expense mutual funds.” However, consistent with our
cognitive aging hypothesis, in untabulated results, we find that older investors do not earn higher
risk-adjusted returns from their increased participation rates in mutual funds. After accounting
for other factors, age is negatively related and experience is positively related to risk-adjusted
mutual fund performance. This evidence shows that in yet another setting older investors follow
the commonly prescribed investment principle but are relatively less effective in applying that
rule.22
The evidence from other related studies reinforce our diversification skill results. In particu-
lar, Bailey, Kumar, and Ng (2008) show that both older and experienced investors exhibit strong
intentions to diversify using foreign stocks, but only the experienced investors benefit from their
diversification attempts. This evidence indicates that older investors are more likely to attempt
to exploit the potential benefits of foreign investments but, conditional upon participation, they
appear less skillful in their decisions.
E. Conditional Deterioration in Stock Selection Abilities
When we estimate the conditional skill regression with the four-factor alpha as the dependent
variable (a measure of stock selection skill), the results are remarkably similar to the Sharpe
ratio regression estimates. These results are reported in Table IV, Panel A, Column (4). We
find that both Age and Investment Experience have significant coefficient estimates, where the
loading on Age is negative (estimate = −0.051, t-stat = −4.735) and the loadings on Investment
Experience and Education are positive (the estimates are 0.020 and 0.014, and the t-statistics
are 2.283 and 2.527, respectively). Further, the Over 70 Dummy has a negative and significant
24
coefficient estimate (estimate = −0.025, t-stat = −2.073). We also find that the Age×Low
Income and Age×Low Education interaction terms have negative coefficient estimates, though
the latter is not statistically significant.
When we use the mean characteristic-adjusted monthly portfolio return to measure risk-
adjusted performance, consistent with our alpha regression estimates, we find that age and
investment experience maintain their opposite signs and the interaction terms have similar
estimates (see column (6)).23 The coefficient estimate of Age is −0.054 with a t-statistic of
−2.945 and coefficient estimate of Investment Experience is 0.027 with a t-statistic of 3.891. In
addition, both Age×Low Income and Age×Low Education interaction terms have the expected
negative and significant coefficient estimates (the estimates are −0.014 and −0.005 with t-
statistics of −1.996 and −2.719, respectively).
While we use robust, clustered standard errors in our skill regressions to account for po-
tential cross-sectional dependence, for additional robustness, we estimate a panel regression
specification using the month-t characteristic-adjusted portfolio return as the dependent vari-
able. As before, we use the Driscoll and Kraay (1998) methodology to account for potential
cross-sectional correlation, serial correlation, and cross-serial correlation. We find that the panel
regression results reported in Column (7) are qualitatively very similar to the cross-sectional re-
gression estimates.
In untabulated results, we find that the coefficient estimates of our control variables have the
expected signs. For instance, in the alpha regression, the Portfolio Dividend Yield has a strongly
negative estimate, which indicates that investors who focus excessively on high dividend yield
stocks have weaker stock selection ability. Nonetheless, these investors are able to reduce the
total risk of their portfolios, thereby increasing the Sharpe ratio (see column (2)). Additionally,
investors who hold larger portfolios exhibit better stock selection ability because portfolio size
is likely to proxy for greater resources and it may also reflect greater investment experience.24
We also find that the coefficient estimate of Retired Dummy is statistically insignificant, which
suggests that the abrupt lifestyle change at the time of retirement does not have an incremental
25
negative effect on investment skill.
The coefficient estimates in the skill regressions are easy to interpret in economic terms.
They allow us to quantify the performance decline that can be attributed solely to the adverse
effects of cognitive aging. For instance, the coefficient estimate of Age in column (4) indicates
that, all else equal, a one standard deviation shift in the age of an investor who does not belong
either to the lower income, lower education, or ethnic minority groups would be associated
with an annual, risk-adjusted performance decline of 0.051 × 12 = 0.612%. This indicates that
when an investor aged 30 becomes older and crosses the retirement age of 65 (a three standard
deviation change in age), she is likely to suffer an annual performance decline of 1.836% on a
risk-adjusted basis.
Examining the race/ethnicity interactions, we find that Age×Hispanic interaction term has
a significantly negative coefficient estimate in all four regression specifications. For instance, in
the alpha regression, the coefficient estimate of the interaction term is −0.034, with a t-stat of
−3.516. In economic terms, this implies that an older investor who also earns lower income and
belongs to the Hispanic ethnic group would experience an annual, risk-adjusted performance
decline of (0.051 + 0.025 + 0.034) × 12 = 1.320%. For an investor with these attributes, a
jump in age from 30 to 65 would correspond to an annual performance decline of 3.960% on a
risk adjusted basis. In contrast, we find that the Age×African American interaction term has
insignificant coefficient estimates.
F. Stock Selection Skill Regression Estimates using Net Returns
Because older investors have lower turnover rates (see Section IV.B), their net performance after
accounting for transaction costs might not be significantly worse than younger investors. Using
the Barber and Odean (2000) method, we account for trading commissions, bid-ask spread, and
the potential market impact of investors’ trades and measure the net portfolio returns of each
investor portfolio. We re-estimate the skill regression using characteristic-adjusted net returns.
The results are reported in Column (8) of Table IV, Panel A.
26
Even when we use net portfolio return to measure skill, Age and Investment Experience
maintain their opposite signs and relative magnitudes. As expected, we find that the adverse
effects of aging are weaker when we use net returns but the performance decline that can be
attributed to aging remain economically significant. For instance, when an investor aged 30
becomes older and crosses the retirement age of 65 (a three standard deviation shift in age), she
is likely to suffer a net annual performance decline of 0.045 × 12 × 3 = 1.620%. If this investor
also earns lower income and belongs to the Hispanic ethnic group, she would experience a net
annual, risk-adjusted performance decline of (0.045 + 0.023 + 0.020) × 12 × 3 = 3.168%.25
Overall, the skill regression estimates support our unconditional hypothesis (H2), which
posits that investment skill increases with experience due to the positive effects of learning, but
declines with age due to the adverse effects of cognitive aging. The evidence also supports our
conditional hypothesis (H2cond), which posits that the decline in skill is steeper among less
educated and less wealthy older investors who belong to minority groups.
G. Cognitive Aging and Learning
Do older investors learn less effectively due to cognitive aging? To gather additional support
for our main hypothesis, we examine whether the speed and effectiveness of learning varies with
age. We embed two interaction variables in one of our skill regression specifications (column
(6) in Table IV, Panel A): Below 30×Low Experience and Over 70×Low Experience. In unt-
abulated results, we find that lack of experience has stronger adverse effects on older investors.
The coefficient estimate for the Over 70×Low Experience dummy is negative and significant
(estimate = −0.041, t-stat = −4.068) while the coefficient estimate of the Below 30×Low Expe-
rience dummy is statistically insignificant (estimate = 0.005, t-stat = 0.546). This evidence is
consistent with our conditional hypothesis and indicates that learning is impaired by the adverse
effects of cognitive aging.
27
H. Sub-Sample Estimates Without the Extreme Performers
While we have used multiple risk-adjusted performance measures to obtain “style adjusted”
portfolio performance, given our relatively short sample size, one might argue that the perfor-
mance differences reflect systematic age-induced style differences rather than the adverse effects
of cognitive aging. For instance, the style or industries (e.g., the technology sector) favored
by investors with higher cognitive abilities might have yielded exceptional returns during the
six-year sample period. Thus, it is possible that older investors do not have poor stock picking
skills but rather the styles chosen by them performed poorly during the sample period due to
chance.
To examine whether the exceptionally superior or poor performance of certain types of stocks
or industries are influencing our estimates, we exclude k% investors from both the top and the
bottom tails of the performance distribution and re-estimate the skill regressions. Even when
we choose k = 5, our estimates remain qualitatively similar to the baseline estimates reported
in Table IV, Panel A, Column (4). For instance, the untabulated results indicate that the
coefficient estimates for Age and Investment Experience are −0.047 (t-stat = −3.031) and 0.035
(t-stat = 4.609), respectively. Additionally, the interaction terms have the expected negative
and significant estimates.
I. Adverse Effects of Cognitive Aging or Cohort Effects?
Studies like ours that examine the effects of age are plagued with the classic age-cohort-period
identification problem (e.g., Deaton (1997), Ameriks and Zeldes (2004)). The main concern is
that in addition to age-induced cognitive aging effects that we are mainly interested in, age
could capture birth cohort or time effects. Because the three effects are strongly correlated, it
is usually difficult to isolate their effects without a data set that tracks the portfolio choices of
the same set of individuals over a very long period of time. Fortunately, time effects are unlikely
to play a significant role during the relatively short six-year sample period. Thus, we largely
assume that time effects are small and do not contaminate our results. We also find that our
28
main results hold for the 1991-93 and 1994-96 sub-samples (see Table VI), which indicates that
time-effects are not the primary drivers of our findings.
Focusing on cohort effects, we conjecture that there are several reasons why cohort effects
are unlikely to explain our findings. First, we use a combination of rules of thumb and skill
regressions to identify the adverse effects of cognitive aging, where our main hypothesis predicts
opposite effects of age in the two contexts. But just like the effects of experience, any cohort-
based hypothesis would predict a similar influence of age in both regressions. Common social
experiences such as the quality of education, socioeconomic environment when growing up, or
common first hand experience of salient events (e.g., the depression or the stock market crash)
that are often associated with cohort effects cannot successfully explain the opposite signs of age
in rules of thumb and skill regressions. Second, as shown in Figure 1, cohort-based explanations
for the abrupt and sudden drop in performance at older age are unlikely to be very convincing.
Third, cohort effects cannot successfully account for the inverted U-shaped relation between age
and investment skill. In contrast, all these findings are strongly consistent with the cognitive
aging hypothesis and reflect the natural outcome of the joint aging and learning processes.
Although our results so far seem inconsistent with cohort-based explanations, we conduct
several tests to directly account for the effects of birth cohorts. In the first test, we follow Poterba
and Samwick (1997) and define cohort-range dummy variables, where we consider five cohort
ranges: Below 35, 35-45, 45-55, 55-65, and above 65. In untabulated results, we find that the age-
range dummies have negative but insignificant coefficient estimates in all our specifications. More
importantly, the coefficient estimates of age, experience, and other interaction variables that
provide evidence in support of our main hypothesis remain significant. In fact, the coefficient
estimate of Age (see column (6) in Table IV, Panel A) becomes stronger (coefficient = −0.082,
t-stat = −3.129).
Next, we perform an individual-level performance comparison and find that the change in
performance between the first and the second halves of the sample period exhibits an inverted
U-shaped pattern (see Figure 1). The older investors experience a greater decline in performance
29
and similar to the age-skill relation, there is a sharp decline at very old age. This evidence is
consistent with the cognitive aging hypothesis and does not suffer from potential contamination
from cohort effects because differencing eliminates the common cohort effects (McKenzie (2006)).
The age-performance differential relation is qualitatively similar when we consider a multi-
variate estimation framework and account for other factors that might influence the sub-period
performance differential. Specifically, we estimate a cross-sectional regression in which the per-
formance differential between the two halves of the sample period is the dependent variable and
the set of independent variables includes the demographic characteristics used in the skill re-
gression. We find that Age and Investment Experience have opposite signs in the skill difference
regression (see Table IV, Panel A, column (9)). This evidence indicates that older investors
experience a decline in performance, while the performance of investors with greater experience
improves.
When we consider a quadratic age term to capture the hump shape of the age-performance
relation, we find qualitatively similar results. These results are reported in Panel B of Table
IV. When the dependent variable is one of the performance measures, both Age and Age2 have
negative and significant coefficient estimates (see columns (1) to (3)). This evidence indicates
that the inverted U-shape relation between age and investment skill is strong and significant
even when we account for other demographic factors and portfolio characteristics that may be
correlated with skill. When the dependent variable is the sub-period performance differential,
we again find that both Age and Age2 have negative and significant coefficient estimates (see
columns (4) and (5)). In both specifications, we also find that the Over 70 Dummy is signifi-
cantly negative, which is consistent with the sharp drop in performance level and performance
differential documented in Figure 1.
To further investigate whether our evidence reflects the adverse effects of cognitive aging
or cohort effects, we identify a scenario where cohort effects do not make a meaningful predic-
tion but the cognitive aging hypothesis makes a sharp prediction. Specifically, we assume that
portfolio size is a proxy for task difficulty and examine the effect of portfolio size on portfolio
30
performance differential that we attribute to cognitive aging. When an investor actively man-
ages a larger portfolio that requires greater attention and cognitive capacity, the cognitive aging
hypothesis predicts that the possibility of making mistakes would be greater among older in-
vestors. Consequently, the performance differential between younger and older investors would
increase with portfolio size. In contrast, there is no obvious birth cohort-based prediction for
the variation in performance differential with portfolio size.
In our last cohort related test, we include Portfolio Size×Age interaction variable in the skill
regression specification (column (6) in Table IV, Panel A). We find that the interaction term
has a significantly negative coefficient estimate (estimate = −0.032, t-stat = −3.987), while the
estimates of other variables in the model remain almost unchanged. Thus, consistent with the
cognitive aging hypothesis, older investors exhibit worse risk-adjusted performance when they
hold larger portfolios that are more difficult to manage.26 This evidence, however, does not have
a clear cohort-based explanation. Overall, the results from birth cohort related robustness tests
indicate that cohort effects are unlikely to successfully explain the observed negative relation
between age and investment skill.
VI. Economic Costs of Cognitive Aging
A. A Portfolio Based Approach
In this section, we quantify the economic costs associated with cognitive aging more accurately
using portfolio-based, time-series tests. An additional advantage of the portfolio-based analy-
sis is that it is insensitive to concerns about potential cross-sectional dependence in portfolio
performance.
We proceed as follows: First, we follow the standard imputation methodology (e.g., Skinner
(1987), Ziliak (1998), Browning and Leth-Petersen (2003)) and use the previously estimated
model of cognitive abilities (see Section V.C and Table V) to obtain an imputed cognitive ability
31
measure for each investor. We regress this imputed cognitive ability measure on investment
experience and obtain a residual cognitive ability measure that is orthogonal to experience.
Next, we sort investors into deciles based on their residual cognitive ability measures, where
group 1 consists of investors with low cognitive ability and group 10 consists of investors with
high cognitive abilities. Last, we compute the average performance of investors in each of the
ten cognitive ability categories and estimate the economic costs of aging.
We find that investors in the lowest cognitive ability decile earn a mean monthly return of
0.984%, investors in the highest cognitive ability decile earn a mean monthly return of 1.264%,
and the annualized performance differential of 3.360% (12×(1.264−0.984)) is statistically signif-
icant (t-stat = 2.181). The performance differentials are of similar magnitudes when we measure
the risk-adjusted performance using the four-factor alpha (annualized performance differential
= 3.240%, t-stat = 2.762) or the characteristic-adjusted returns (annualized performance differ-
ential = 3.363%, t-stat = 2.827). These performance estimates indicate that the economic costs
of cognitive aging are significant.
Examining the factor exposures of the cognitive ability sorted portfolios, we find that high
cognitive ability investors hold relatively riskier, smaller, and growth-oriented portfolios. For
instance, the four factor exposures (RMRF , SMB, HML, and UMD) for the lowest cognitive
ability decile portfolio are 1.083, 0.571, 0.314, and −0.218, respectively. In contrast, for the
highest cognitive ability decile portfolio, the corresponding factor exposures are 1.219, 0.677,
0.112, and −0.299, respectively. These factor exposure estimates are statistically significant and
the factor estimate differences between the high and the low cognitive ability decile portfolios
are also significant. This evidence is consistent with our evidence on investors’ stock preferences,
conditional upon age (see Section IV.A and Table II).
To examine whether the economic costs of aging are also economically significant for in-
vestors who hold larger portfolios, we compute the risk-adjusted performance measures, con-
ditional upon portfolio size. The results are shown in Figure 2. We find that the economic
cost of cognitive aging increases with portfolio size. As portfolio size increases, high (quintile 5)
32
cognitive ability investors perform better, low (quintile 1) cognitive ability investors do worse,
and the performance gap becomes wider. When portfolio size is below $10,000, the annualized
characteristic-adjusted return differential is 1.725% and it increases to 5.012% when the portfo-
lio size is above $50,000. When we use the four-factor alpha as the performance measure, the
cognitive aging related performance differentials for small, mid-sized, and large portfolios are
2.148%, 2.407%, and 4.613%, respectively.27
Collectively, the results from portfolio-based time-series tests indicate that investors who
do not experience the adverse effects of cognitive aging earn superior risk-adjusted returns.
Importantly and somewhat surprisingly, the economic costs of cognitive aging are larger (both
in percentage and dollar terms) among investors who hold larger portfolios. Those investors are
less likely to hold significant positions outside their brokerage accounts, especially during the
sample period when the median number of brokerage accounts varied between one and two.28
The performance differentials measured in percentage terms might not appear alarming, but
because older investors hold larger portfolios, the economic costs are even more significant when
measured in dollar terms.
B. Identifying the Components of Stock Selection Ability
To better understand how high cognitive ability investors generate superior portfolio perfor-
mance, we use the Daniel, Grinblatt, Titman, and Wermers (1997) decomposition to estimate
the three components of stock selection skill: characteristic selectivity (CS), characteristics tim-
ing (CT ), and average style (AS). A positive estimate for CS reflects stock selection ability
within the style portfolios, while a positive CT estimate provides evidence of style timing.
First, we sort investors into five or ten categories based on their residual cognitive ability
measures and construct an aggregate portfolio for each investor category by combining the
portfolios of all investors who belong to the group. Then, for each of those aggregate portfolios,
we compute the CS, CT , and AS measures. We find that high cognitive ability investors exhibit
greater ability to pick stocks within the styles. When we consider five portfolios, the annual CS
33
measures for the lowest and the highest cognitive ability quintile portfolios are −0.496% and
0.660%, respectively, and the difference of 1.156% is significant. As expected, the performance
differential is stronger (= 1.920%) when we consider cognitive ability sorted decile portfolios.
Examining the CT estimates for the cognitive ability sorted portfolios, we find that CT
estimates are uniformly negative for all portfolios. Thus, both high and low cognitive ability
investors lack characteristic timing abilities. However, the low cognitive ability investors have
more negative CT measure and exhibit worse timing abilities. For instance, the annual CT
measure for the lowest and the highest decile portfolios are −2.001% and −1.156%, respectively,
and the difference is positive (= 0.845%).
The AS estimates for the cognitive ability sorted portfolios exhibit less variation and are
similar. Nevertheless, the performance difference between the highest and the lowest cognitive
ability categories is positive. When we consider quintile portfolios, the annual AS differential
is 0.292% and when we consider decile portfolios, the differential is 0.429%. Overall, the per-
formance differentials between the highest and the lowest cognitive ability sorted quintile and
decile portfolios are 1.949% and 3.195%, respectively.
VII. Alternative Explanations and Robustness Checks
In this section, we conduct a wide array of tests to examine the robustness of our results
and attempt to further rule out alternative interpretations for our findings. The results are
summarized in Table VI, where for brevity, we only report the estimates for the six variables
that are most closely related to our aging related hypotheses.
A. Alternative Skill Measures
In the first set of tests, we consider other investment skill measures. First, we compute a trading-
based skill measure that is not strongly correlated with our previous performance-based skill
measures and does not depend upon our ability to observe investors’ entire financial portfolios.
34
This skill measure reflects the belief that investors who have superior stock selection ability are
likely to buy stocks that perform better than the stocks they sell. Our choice is motivated by
Odean (1999) and Barber and Odean (2001), who use a similar measure to identify the stock
selection ability of investors.
Specifically, we use the mean k-day post-trade buy-sell return differential (PTBSD(k)) as a
measure of investment skill. We choose two different values of k (k = 5 and 10), because using the
same dataset as ours, Coval, Hirshleifer, and Shumway (2005) show that the trading performance
of individual investors declines after about two weeks. When PTBSD(5) or PTBSD(10) is
used as a dependent variable in the skill regression, similar to the results from previous skill
measures, we find that the Age and Investment Experience have opposite signs (see tests (1) and
(2) in Table VI). The interaction terms mostly have the expected signs, though their estimates
are statistically weaker. This evidence indicates that older investors trade less (see Table III),
but when they do trade, they make worse buy and sell decisions. The stocks they purchase
under-perform the stocks they sell by a larger magnitude.
Next, we examine investors’ market timing abilities using the two Graham-Harvey perfor-
mance metrics (Graham and Harvey (1996, 1997)). We re-estimate the skill regression and
find that, in both instances, age and investment experience maintain their opposite signs (see
tests (3) and (4)). In addition, the interaction term estimates are very similar to the respective
estimates in the alpha skill regression reported in Table IV, Column (4). These results are
consistent with our second main hypothesis and indicate that market timing skill improves with
experience but deteriorates with age due to the adverse effects of cognitive aging.
B. An Alternative Experience Measure
In our second robustness test, we define an alternative experience measure using our main
investment experience proxy (number of days since account opening date) and self-reported
measures of investment knowledge and experience. For a small sub-set of investors, we know
whether investors believe that they have extensive, good, limited, or zero investment experience.
35
A similar self-reported measure reflects their self-assessment of investment knowledge. Using
these two self-reported measures, we construct an indicator variable that is set to one for investors
who have either good or extensive investment knowledge or experience. We also define a second
indicator variable that is set to one for investors who hold the brokerage account for at least
five years. Our new experience measure is the sum of the two indicator variables.29
When we replace the investment experience proxy in the skill regression with this new expe-
rience variable, we still find that Age has a significantly negative coefficient estimate (coefficient
= −0.054, t-stat = −3.690) and Experience has a significantly positive coefficient estimate (co-
efficient = 0.033, t-stat = 2.390). The other main variables in the skill regression specification
maintain their original signs and significance levels (see test (5)). This evidence indicates that
our key results are robust and do not depend upon the specific choice of the investment experi-
ence proxy.
C. Aging Effects or Informational Illiteracy?
In the third robustness test, we entertain another cohort-based explanation for the negative age-
skill relation. Specifically, we examine whether older investors make worse investment decisions
because they are slower to adopt new technology such as the Internet and not due to cognitive
aging. It is possible that older investors who belong to a cohort from the “Radio and Television”
era and are not as technologically savvy as the cohort of younger investors from the “Computers
and Internet” era. Consequently, older investors could experience higher search costs and make
their decisions using stale and less accurate information signals.
We consider a subset of investors who are technologically savvy and have used the Internet
to execute trades in their brokerage account. If the adverse effects of aging are visible even
among this set of investors with relatively higher levels of informational literacy, it is unlikely
that cohort induced differences in technology adoption propensity can successfully explain our
findings. When we use the four-factor alpha skill measure to estimate the skill regression using
the sub-sample of Internet users only, we find that Age and Investment Experience maintain
36
their opposite signs (see test (6)). Age has a coefficient estimate of −0.063 (t-stat = −3.168)
and Experience has a coefficient estimate of 0.048 (t-stat = 2.653). The other main variables
in the skill regression specification maintain their original signs and significance levels. These
results are qualitatively similar to our baseline estimates reported in Table IV and indicate that
informational literacy associated with an age cohort is unlikely to be an appropriate explanation
for our findings.
D. Split Sample Tests
In our fourth robustness test, we re-estimate the skill regression for two sub-periods. The
1991 to 1996 sample period encapsulates two distinct market conditions: (i) a relatively flat
market during the first half of the sample period, following the 1990 NBER recession, and (ii)
an increasing market during the second half of the sample period, representing the start of the
“bubble” period. One might be concerned that the negative age-skill relation we find results from
the relatively conservative (i.e., less risky) investments by older investors during bullish market
conditions. To examine whether the adverse effects of cognitive aging are significant during
both periods of weak and strong stock market performance, we estimate the skill regression for
the 1991 to 1993 and the 1994 to 1996 sub-periods using the characteristic-adjusted portfolio
returns. We use the characteristic-adjusted performance measures because the four-factor alpha
estimates would be noisier with only three years of data.
The skill regression estimates for the two sub-samples indicate that the adverse effects of
cognitive aging and the positive effects of learning are present in both time-periods (see tests (7)
and (8)). Both age and experience variables have significant coefficient estimates. Interestingly,
the age-race/ethnicity interaction terms have statistically weak coefficient estimates in the full-
sample regression but they have the expected negative signs during the 1994 to 1996 sub period.
Taken together, the sub-period estimates indicate that the negative age-skill relation we find
cannot be attributed to the bullish market conditions during the latter part of the sample
period.
37
E. Differential Skill in Identifying Superior Local Stocks
The fifth robustness test focuses on the performance of local investments. Ivkovic and Weisben-
ner (2005) show that, on average, the local stock investments of individual investors perform
better than their non-local investments. To examine whether the negative age-skill relation hold
in both local and non-local settings, we compute the four-factor alpha for each investor’s local
portfolio. The local portfolio represents the part of the portfolio that contains stock investments
in firms located within a 100 mile radius of investor’s location.30 We re-estimate the skill re-
gressions using the local alpha measure as the dependent variable. The correlation between the
local and the total alpha measures is positive (= 0.513) but not very high.
The skill regression estimates are again consistent with our second main conjecture. We find
that age and investment experience estimates have opposite signs, and the age-race/ethnicity
interaction terms have negative coefficient estimates (see test (9)). The Age×Low Income and
Age×Low Education interaction terms have the expected negative signs but those estimates are
statistically insignificant. This evidence indicates that older investors experience the adverse
effects of cognitive aging even in their local stock investment decisions.
F. Controls for Liquidity and Industry Exposures
In the next robustness test, we further ensure that our results are not influenced by the ex-
ceptional performance of certain styles or industries during the relatively short six-year sample-
period. We follow the Pastor and Stambaugh (2002) and Pastor and Stambaugh (2003) method-
ologies and compute portfolio alphas after employing controls for industry exposures and liquid-
ity. We estimate skill regressions using this eight-factor alpha and find that Age is still negatively
related to investment skill (estimate = −0.047, t-stat = −3.745), while Investment Experience
is still positively related to investment skill (estimate = 0.022, t-stat = 2.314). Furthermore,
the interaction terms maintain their negative coefficient estimates (see test (10)).
For additional robustness, we follow an alternative approach to control for the effects of indus-
tries. For each investor, we compute the mean portfolio weight allocated to the 48 Fama-French
38
industries (Fama and French (1997)) and employ them as additional independent variables in the
skill regressions. In untabulated results, we find that many of these industry weights have large
and significant coefficient estimates. For instance, technology stocks in the electronic equipment
industry has a strong positive coefficient estimate (estimate = 0.114, t-stat = 10.254) while util-
ities industry has a strong negative coefficient estimate (estimate = −0.152, t-stat = −12.773).
However, the age, experience, and relevant interaction variables still maintain their signs and
statistical significance. Thus, the exceptional performance of one or more industries during the
sample period does not significantly influence our skill regression estimates.
G. Lack of Skill or Lack of Effort?
The seventh robustness test examines the “play money” hypothesis. One might argue that
the negative skill-aging relation we find is not too surprising or economically very meaningful
because the portfolios we analyze represent investors’ play money accounts meant primarily for
gambling and entertainment purposes. The bulk of investors’ actual investments including their
retirement accounts are held elsewhere, which we cannot observe.
To examine whether the weaker investment skill of older investors can be attributed to their
lack of interest in their brokerage portfolios, we consider a sub-sample of investors who hold
larger portfolios in comparison to their income levels. Specifically, we use the four-factor alpha
skill measure and re-estimate the skill regressions for the sub-sample of investors whose mean
portfolio size to annual income ratio (i.e., SIR) is greater than or equal to 1.23 (the top two-third
of investors). These equity portfolios are unlikely to represent investors’ play money accounts
and are less likely to be ignored.
The skill regression estimates for the large portfolio sub-sample indicate that the sub-sample
coefficient estimates are similar to the full sample results (see test (11)). For instance, Age has
a coefficient estimate of −0.053 (t-stat = −4.759) and Investment Experience has a coefficient
estimate of 0.018 (t-stat = 1.996) in the alpha regression. Additionally, two of the four interac-
tion terms have the expected negative signs. Thus, our key results are unlikely to be successfully
39
explained by the play money hypothesis.
H. Are the Results Driven by Investors Located in California?
In the eighth robustness test, we examine whether our results are influenced by the strong
concentration of investors from California because a considerable portion (27.25%) of our sample
is located in California. To guard against this possibility, we exclude all investors who reside
in California and re-estimate the skill regression using the four-factor alpha skill measure. The
results for the non-Californian sub-sample are similar to the full sample results (see test (12)).
Age has a coefficient estimate of −0.044 (t-stat = −2.797) and Investment Experience has a
coefficient estimate of 0.017 (t-stat = 2.651) in the alpha regression. This result indicates that
the geographical concentration of our sample investors is unlikely to successfully explain the
negative age-skill relation.
I. Is the Age on the Account Opening Date An Indicator of Skill?
In the last robustness test, we examine whether our main results are influenced by a sample
selection bias. In particular, one might be concerned that investor’s age on account opening
date might be an indicator of skill. All else equal, investors who open the brokerage account at
an older age would have lower investment skill. Thus, the investment skill differences between
older and younger investors after accounting for experience might not reflect the adverse effects
of cognitive aging but could merely reflect the sample selection bias.
To ensure that our results not influenced by the potential sample selection bias, we compute
the age of each investor on the account opening date (= Age in June 1997 − Investment Expe-
rience) and estimate skill regressions for younger, middle-aged, and older investors sub-samples.
We use the age on the account opening date to define the three sub-samples. In untabulated
results, we find that in both diversification skill and stock selection skill regressions, the age and
experience variables maintain their opposite signs and magnitudes. This evidence indicates that
40
our results do not reflect skill differences that can be attributed to investor’s age on account
opening date.
Collectively, the additional robustness test results indicate that the empirical support for the
unconditional (H2) and conditional (H2cond) hypotheses is strong and the results are remarkably
consistent with the extant psychological evidence. Our findings are not sensitive to the choice of
skill measures, potential error in skill measurement, choice of the risk adjustment methodology,
specific market conditions, investors’ lack of interest in relatively small brokerage portfolios, and
the geographical concentration of our sample investors.
VIII. Summary and Implications of Our Research
This is the first study to examine the potential role of cognitive aging on the stock investment
decisions of older investors. We investigate whether older individual investors make better
investment choices because of greater investment experience or whether their investment skill
deteriorates with age due to the adverse effects of cognitive aging. This is an important issue
that has implications for how individual investors should structure their portfolios over time,
the type of investment advice they should seek over their lifetime, and the potential effects of
changes in government policy on investment generated retirement income.
Our evidence indicates that older and more experienced investors hold less risky portfolios,
exhibit stronger preference for diversification, trade less frequently, exhibit greater propensity for
year-end tax-loss selling, and exhibit weaker behavioral biases such as the disposition effect and
familiarity bias. Thus, their choices reflect greater knowledge about investing. But consistent
with the cognitive aging hypothesis, we also find that older investors have worse investment
skill, where the skill deteriorates sharply around the age of 70. Examining the economic costs
of aging, we find that older investors earn about 3-5% lower annual return on a risk-adjusted
basis. Collectively, our evidence indicates that older investors’ portfolio choices reflect greater
knowledge about investing but their investment skill deteriorates with age due to the adverse
41
effects of cognitive aging.
These results could potentially be used to improve the investment decisions of older investors.
We do not prescribe that older investors should stop making independent investment decisions.
Rather, based on the evidence, our hope is that older people would recognize the adverse effects
of cognitive aging and would try to compensate for those effects, perhaps by seeking advice
from a financial advisor or some other qualified investment professional. An investment refresher
course that highlights the adverse effects of cognitive aging and prescribes effective compensating
mechanisms might also be helpful. This may especially be a wise decision for investors who hold
portfolios larger than those examined in this study. Our performance results also indicate that
while most investors in our sample would benefit from holding a passive index fund, the potential
benefits of passive investing are likely to be higher for older investors.
In broader terms, our empirical findings could help us better understand the stock market
participation puzzle. Theoretical models typically have the greatest difficulty in explaining the
participation rates in the extreme age categories (e.g., Gomes and Michaelides (2005)). Based on
our findings, we conjecture that younger investors would stay away from the stock market due to
their lack of investment experience, while older investors would be less willing to participate due
to a perception of declining cognitive abilities. A theoretical model that synthesizes the positive
effects of experience and adverse effects of cognitive aging may provide a fresh perspective into
the stock market participation and the broad asset allocation decisions over the life-cycle.
Our empirical evidence also provides specific guidance for refining asset pricing models that
incorporate the effects of aging. Previous theoretical models have examined the aggregate effects
of aging on the stock market behavior (e.g., Bakshi and Chen (1994), Poterba (2001), Goyal
(2004)) through the channel of risk aversion. Our results indicate that age is likely to influence
asset returns through an additional channel. Specifically, if older investors become aware of their
declining investment skill, the perceived costs for stock market participation would increase,
and those investors would demand a higher premium for investing in the stock market. In
this scenario, stock market returns could increase as the population ages, even when there is no
42
substantial increase in people’s risk aversion. Additionally, our evidence on the stock preferences
of older investors provide guidance about the segments of the market (e.g., high dividend yield
and less risky stocks), where the effects of aging on stock returns are likely to be stronger.
We conclude the paper with a caveat. Although our results are strongly consistent with
the cognitive aging hypothesis, some amount of caution must be exercised while interpreting
this evidence because we cannot directly measure the degree of cognitive decline among older
investors. In addition, because we use a relatively short six-year panel to identify the adverse
effects of cognitive aging, in spite of our numerous attempts to rule out alternative explanations
for our results based on cohort and time effects, some concerns might remain. Nevertheless, given
the remarkable similarities between our results and the evidence from psychological research on
aging, the footprints of cognitive aging in investment decisions appear strong and difficult to
ignore.
43
References
Agarwal, Sumit, John C. Driscoll, Xavier Gabaix, and David Laibson, 2007, The age of reason:
Financial decisions over the life cycle, Working Paper (March), Harvard University and MIT.
Ameriks, John, and Stephen Zeldes, 2004, How do household portfolios shares vary with age?,
Working Paper (September), Graduate School of Business, Columbia University.
Arbuckle, Tannis Y., Dolores Gold, and David Andres, 1986, Cognitive functioning of older
people in relation to social and personality variables, Journal of Psychology and Aging 1,
55–62.
Avolio, Bruce J., and David A. Waldman, 1986, Variation in cognitive, perceptual and psy-
chomotor abilities across the working life span: Examining the effects of race, sex, experience,
education, and occupational type, Journal of Psychology and Aging 1, 55–62.
Bailey, Warren, Alok Kumar, and David Ng, 2008, Foreign investments of U.S. individual in-
vestors: Causes and consequences, Management Science 54, 443–459.
, 2009, Behavioral biases and mutual fund clienteles, Working Paper (April), Cornell
University and University of Texas at Austin.
Bakshi, Gurdip S., and Zhiwu Chen, 1994, Baby boom, population aging, and capital markets,
Journal of Business 67, 165–202.
Baltes, Margaret M., and Frieder R. Lang, 1997, Everyday functioning and successful aging:
The impact of resources, Psychology and Aging 12, 433–443.
Baltes, Paul B., and Ulman Lindenberger, 1997, Emergence of a powerful connection between
sensory and cognitive functions across the adult life span: A new window to the study of
cognitive aging?, Psychology and Aging 12, 12–21.
Barber, Brad M., and Terrance Odean, 2000, Trading is hazardous to your wealth: The common
stock investment performance of individual investors, Journal of Finance 55, 773–806.
44
, 2001, Boys will be boys: Gender, overconfidence, and common stock investment,
Quarterly Journal of Economics 116, 261–292.
Black, Sheila, 2004, Cognitive functioning among African American older adults, African Amer-
ican Research Perspectives 10, 106–118.
Browning, Martin, and Soren Leth-Petersen, 2003, Imputing consumption from income and
wealth information, Economic Journal 113, F282–301.
Cagney, Kathleen A., and Diane S. Lauderdale, 2002, Education, wealth, and cognitive function
in later life, Journal of Gerontology 57B, 163–172.
Calvet, Laurent E., John Y. Campbell, and Paolo Sodini, 2007, Down or out: Assessing the
welfare costs of household investment mistakes, Journal of Political Economy 115, 707–747.
Campbell, John Y., 2006, Household finance, Journal of Finance 61, 1553–1604.
, and Luis M. Viceira, 2002, Strategic Allocation: Portfolio Choice for the Long-Term
Investor (Oxford University Press, New York, NY).
Chevalier, Judith, and Glenn Ellison, 1999, Are some mutual fund managers better than others?
Cross-sectional patterns in behavior and performance, Journal of Finance 54, 875–899.
Christelis, Dimitris, Tullio Jappelli, and Mario Padula, 2007, Cognitive abilities and portfolio
choice, CEPR Working Paper # 5735.
Cocco, Joao F., Francisco J. Gomes, and Pascal J. Maenhout, 2005, Consumption and portfolio
choice over the life cycle, Review of Financial Studies 18, 491–533.
Constantinides, George M., 1983, Capital market equilibrium with personal tax, Econometrica
51, 611–636.
, 1984, Optimal stock trading with personal taxes: Implications for prices and the
abnormal January returns, Journal of Financial Economics 13, 65–89.
45
Coval, Joshua D., David Hirshleifer, and Tyler Shumway, 2005, Can individual investors beat
the market?, Working Paper # 04-025, Harvard Business School.
Craik, Fergus I. M., and Timothy A. Salthouse, 1992, Handbook of Aging and Cognition (Erl-
baum, Hillsdale, NJ).
Daniel, Kent D., Mark Grinblatt, Sheridan Titman, and Russell Wermers, 1997, Measuring
mutual fund performance with characteristic-based benchmarks, Journal of Finance 52, 1035–
1058.
Deaton, Angus, 1997, The Analysis of Household Surveys: A Microeconometric Approach to
Development Policy (The Johns Hopkins University Press, Baltimore, MD).
Dhar, Ravi, and Ning Zhu, 2006, Up, close and personal: An individual level analysis of the
disposition effect, Management Science 52, 726–740.
Driscoll, John C., and Aart C. Kraay, 1998, Consistent covariance matrix estimation with spa-
tially dependent panel data, Review of Economics and Statistics 80, 549–560.
Fair, Ray C., 1994, How fast do old men slow down?, Review of Economics and Statistics 76,
103–118.
, 2004, Estimated age effects in athletic events and chess, Experimental Aging Research
33, 37–57.
Fama, Eugene F., and Kenneth R. French, 1997, Industry costs of capital, Journal of Financial
Economics 43, 153–193.
Feng, Lei, and Mark S. Seasholes, 2005, Do investor sophistication and trading experience
eliminate behavioral biases in financial markets?, Review of Finance 9, 305–351.
Goetzmann, William N., and Alok Kumar, 2008, Equity portfolio diversification, Review of
Finance 12, 433–463.
46
Goetzmann, William N., Lingfeng Li, and K. Geert Rouwenhorst, 2005, Long-term global market
correlations, Journal of Business 78, 1–38.
Gomes, Francisco, and Alexander Michaelides, 2005, Optimal life-cycle asset allocation: Under-
standing the empirical evidence, Journal of Finance 60, 869–904.
Goyal, Amit, 2004, Demographics, stock market flows and stock returns, Journal of Financial
and Quantitative Analysis 39, 115–142.
Grady, Cheryl L., and Fergus I. M. Craik, 2000, Changes in memory processing with age, Current
Opinion in Neurobiology 10, 224–231.
Graham, John R., and Campbell R. Harvey, 1996, Market timing ability and volatility implied
in investment newsletters’ asset allocation recommendations, Journal of Financial Economics
42, 397–421.
, 1997, Grading the performance of market timing newsletters, Financial Analysts Jour-
nal 53, 54–66.
Graham, John R., and Alok Kumar, 2006, Do dividend clienteles exist? Evidence on dividend
preferences of retail investors, Journal of Finance 61, 1305–1336.
Horn, John L., 1968, Organization of abilities and the development of intelligence, Psychological
Review 75, 242–259.
Ivkovic, Zoran, James Poterba, and Scott Weisbenner, 2005, Tax-motivated trading by individ-
ual investors, American Economic Review 95, 1605–1630.
Ivkovic, Zoran, Clemens Sialm, and Scott Weisbenner, 2008, Portfolio concentration and the
performance of individual investors, Journal of Financial and Quantitative Analysis 43, 613–
656.
Ivkovic, Zoran, and Scott Weisbenner, 2005, Local does as local is: Information content of the
geography of individual investors’ common stock investments, Journal of Finance 60, 267–306.
47
Kumar, Alok, 2009, Who gambles in the stock market?, Journal of Finance Forthcoming.
, and Sonya S. Lim, 2008, How do decision frames influence the stock investment choices
of individual investors?, Management Science 54, 1052–1064.
Lindenberger, Ulman, and Paul B. Baltes, 1994, Sensory functioning and intelligence in old age:
A strong connection, Psychology and Aging 9, 339–355.
, 1997, Intellectual functioning in old and very old age: Cross-sectional results from the
Berlin aging study, Psychology and Aging 12, 410–432.
List, John A., 2003, Does market experience eliminate market anomalies?, Quarterly Journal of
Economics 118, 41–71.
Masunaga, Hiromi, and John Horn, 2001, Expertise and age-related changes in components of
intelligence, Psychology and Aging 16, 293–311.
McKenzie, David J., 2006, Disentangling age, cohort, and time effects in the additive model,
Oxford Bulletin of Economics and Statistics 68, 473–495.
Merton, Robert C., 1969, Lifetime portfolio selection under uncertainty: The continuous time
case, Review of Economics and Statistics 51, 247–257.
Mossin, Jan, 1968, Optimal multiperiod portfolio policies, Journal of Business 41, 215–229.
Odean, Terrance, 1999, Do investors trade too much?, American Economic Review 89, 1279–
1298.
Pastor, Lubos, and Robert F. Stambaugh, 2002, Mutual fund performance and seemingly unre-
lated assets, Journal of Financial Economics 63, 315–349.
, 2003, Liquidity risk and expected stock returns, Journal of Political Economy 111,
642–685.
48
Poterba, James M., 2001, Demographic structure and asset returns, Review of Economics and
Statistics 83, 565–584.
, and Andrew A. Samwick, 1997, Household portfolio allocation over the life cycle, NBER
Working Paper 6185.
Salthouse, Timothy A., 2000, Aging and measures of processing speed, Biological Psychology
54, 35–54.
Samuelson, Paul A., 1969, Lifetime portfolio selection by dynamic stochastic programming,
Review of Economics and Statistics 51, 239–243.
Schroeder, David H., and Timothy A. Salthouse, 2004, Age-related effects on cognition between
20 and 50 years of age, Personality and Individual Differences 36, 393–404.
Shanan, J., and R. Sagiv, 1982, Sex differences in intellectual performance during middle age,
Human Development 25, 24–33.
Skinner, Jonathan, 1987, A superior measure of consumption from the panel study of income
dynamics, Economic Letters 23, 213–216.
Spaniol, Julia, and Ute J. Bayen, 2005, Aging and conditional probability judgements: A global
matching approach, Psychology and Aging 20, 165–181.
Ziliak, James P., 1998, Does the choice of consumption measure matter? An application to the
permanent-income hypothesis, Journal of Monetary Economics 41, 201–216.
49
Notes
1According to the 2004 Federal Interagency Forum on Aging-Related Statistics, older people (people aged
65 and above) represented about 12% of the population in 2000, but by 2030, their proportion is expected to
increase to about 20%. During the same period, the proportional share of African Americans and Hispanics in
the older population is expected to increase from 14% to 31%.
2The brokerage account opening date can be prior to the sample start date of January 1991. The earliest
accounts were opened in 1972 and all investors in the sample have an account opening date prior to 1992. About
88% of investors in the sample opened their accounts prior to the beginning of the sample period.
3In a related study, Agarwal, Driscoll, Gabaix, and Laibson (2007) examine whether the interaction between
cognitive abilities and experience generates a hump-shaped pattern in financial sophistication, which influences
the prices people pay for financial services.
4Some theoretical models (e.g., Mossin (1968), Samuelson (1969), Merton (1969)) also predict that portfolio
holdings of investors would remain constant with age. See Ameriks and Zeldes (2004) for an excellent review of
this literature.
5There is an interesting (though imperfect) analogy between the investment behavior of older investors we are
trying to capture and the driving behavior of older people. Older drivers may accumulate more driving knowledge
because of their greater driving experience, but they may not be able to apply that knowledge effectively due
to a decline in their physical abilities. In a similar manner, older investors would have greater knowledge
about investing, but they may still fail to apply them effectively due to decline in their general intelligence and
information processing abilities. The analogy suggests that just as older drivers face additional aging related
risks on the road, older investors are likely to expose their portfolios to non-compensated risks due to the adverse
effects of cognitive aging.
6Because the demographic information is available for only a subset of investors in the sample (e.g., both age
and income measures are available for only 31,260 investors), the number of observations in our cross-sectional
regressions depends upon the subset of demographic variables included. See Barber and Odean (2001, Section
II.A) for additional details about the available demographic information.
7See Ivkovic, Poterba, and Weisbenner (2005) and Ivkovic, Sialm, and Weisbenner (2008) for additional
50
discussions on the representativeness of the brokerage data.
8The U.S. Census data are available at http://www.census.gov/main/www/cen1990.html. The SHARE
data are described in Christelis, Jappelli, and Padula (2007) and are available at http://www.share-project.org/.
Professor Kenneth French’s data library is available at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/.
Professor Russ Wermer’s data library is available at http://www.smith.umd.edu/faculty/rwermers/ftpsite/Dgtw/coverp
9The excess portfolio weight allocated to stock i in month t is given by: EW ipt =wipt−wimt
wimt× 100, where
wipt is the actual weight assigned to stock i in group portfolio p in month t and wimt is the weight of stock i in
the aggregate market portfolio in month t.
10The results are remarkably similar when we use other age categories. Specifically, we examined the robust-
ness of our results using the following three age groups: (i) 20-40 (weak retirement motive), (ii) 41-65 (strong
retirement motive), and (iii) above 65 (retired).
11Optimal trading behavior in presence of taxes requires skill (e.g., Constantinides (1983, 1984), Ivkovic,
Poterba, and Weisbenner (2005)). However, such optimal response to taxes would lead to higher risk-adjusted
performance, which our risk-adjusted performance measures are likely to capture (see Section V).
12We obtain similar results when we use other measures (e.g., normalized portfolio variance) to capture the
preference for diversification. Also, see the results in Table V of Goetzmann and Kumar (2008).
13The monthly portfolio turnover rate is the average of purchase and sales turnover rates. The purchase
turnover rate in month t is the ratio of the dollar value of purchases in month t (beginning of month stock prices
are used to compute the value) and the dollar value of the portfolio at the end of month t−1. The sales turnover
rate is defined in an analogous manner.
14In untabulated results, we find that both older and experienced investors are more likely to invest in mutual
funds. However, older investors exhibit a greater propensity to invest in index funds, while experienced investors
exhibit a stronger preference for other types of funds.
15See Bailey, Kumar, and Ng (2008) for additional results on the relation between age and foreign stock
holdings. Bailey, Kumar, and Ng (2009) provide additional evidence on the effect of age on the mutual fund
choices of individual investors.
16The disposition effect evidence is also reported in Dhar and Zhu (2006), although they do not focus on the
51
opposite effects of cognitive aging and investment experience.
17The cognitive ability index is constructed using self-reported measures of ability on a five point scale. Specif-
ically, the Cognitive Ability Index = (Memory − Do Not Get Abstract Idea + Rich Vocabulary + Use Difficult
Words)/4. It is positively correlated with the experience measure but negatively correlated with age.
18Goetzmann, Li, and Rouwenhorst (2005) show that the total portfolio variance can be reduced by increasing
the number of stocks in the portfolio and by a proper selection of stocks such that the average correlation among
stocks in the portfolio is lower. Variance reduction through proper stock selection reflects “skill” while addition
of stocks in the portfolio without a reduction in the average portfolio correlation reflects a “passive” form of
diversification.
19Even when we consider other performance benchmarks that include industry and liquidity factors, our key
results remain qualitatively similar. See Section VII.F.
20Consistent with our evidence, using household-level data from Sweden, Calvet, Campbell, and Sodini (2007)
show that older investors make cautious but relatively less efficient investment choices. Also, see Campbell
(2006).
21In a different context, Chevalier and Ellison (1999) find that, keeping experience and other characteristics
constant, older mutual fund managers exhibit worse performance than younger managers. They find this evidence
puzzling and attribute it to managers’ career concerns. However, the evidence is also consistent with our
unconditional hypothesis, which posits that skill varies inversely with age and positively with experience.
22Also, see Bailey, Kumar, and Ng (2009) for additional evidence on a negative relation between age and
mutual fund performance.
23The correlation between the four-factor alpha and the characteristic-adjusted performance measures is pos-
itive (0.592) but not extremely high.
24To eliminate potential concerns about reverse causality (portfolio size is larger because of better portfolio
performance and not vice versa), we use the portfolio size from the first month the investor enters the sample.
25The results are very similar when the skill measure is the four-factor alpha computed using net returns. The
coefficient estimate of Age is −0.057 (t-stat = −5.366) while the coefficient estimate of Investment Experience is
0.030 (t-stat = 4.380). The interaction terms have the expected signs too. For brevity, we do not tabulate those
52
results.
26Also, see Figure 2. We find similar results when we use the number of stocks in the portfolio to quantify the
difficulty level of managing a portfolio. The coefficient estimate is −0.030 and t-stat = −3.008.
27As before, we find qualitatively similar results when we use the number of stocks in the portfolio to quantify
the difficulty associated with managing a portfolio.
28According to the 1992 Survey of Consumer Finances (SCF), the median U.S. household held only one
brokerage account (mean = 1.57) in 1992 (about 62% of households had only one brokerage account) and the
1995 SCF indicates that the median number of brokerage accounts increased to two (mean = 2.62) in 1995.
29We find similar results if we define the new experience measure as the interaction between the two indicator
variables.
30Following Ivkovic and Weisbenner (2005), we choose the 100-mile cutoff to define local and non-local portfolio
components, but the results are similar when we use a 250-mile cutoff.
53
TABLE I. – SUMMARY STATISTICS: AGE SORTED INVESTOR GROUPS
Age Group Portfolio Size Income Wealth SIR SWR Experience
Panel A: All Investors
Q1 (20-38) $23,372 $90,146 $196,765 0.414 0.283 5.46
Q2 (39-46) $23,372 $98,619 $243,081 0.377 0.246 7.68
Q3 (47-52) $29,270 $101,086 $247,701 0.462 0.304 8.92
Q4 (53-59) $32,543 $93,363 $296,595 0.554 0.298 9.01
Q5 (60-94) $49,274 $70,700 $360,403 1.096 0.346 10.49
Mean $31,566 $90,782 $268,909 0.581 0.295 8.71
Panel B: Investors With Age 60 or Above
Q1 (60-62) $37,652 $87,134 $340,955 0.599 0.293 9.85
Q2 (63-64) $39,678 $74,890 $356,931 0.765 0.322 10.39
Q3 (65-70) $48,416 $72,833 $409,153 1.040 0.313 10.33
Q4 (71-74) $48,556 $67,574 $350,882 0.984 0.338 10.45
Q5 (75-94) $64,526 $64,111 $334,917 1.469 0.464 10.87
Portfolio size is the mean portfolio size of the investors in the group during the sample period. Income is the
annual household income and wealth is the self-reported net worth reported at the account opening date. SIR
is the portfolio size to income ratio, and SWR is the portfolio size to wealth ratio. Investment experience is the
number of years between the account opening date and December 31, 1996. In Panel A (Panel B), there are
5,404 (1,351) observations in each group.
54
TABLE II. – AGE AND STOCK PREFERENCES: PANEL REGRESSION ESTIMATES
Age Groups
Variable (1): 20-38 (2): 60-94 (3): 75-94 (2)−(1) (3)−(1)
Mean Return −0.118 −0.146 −0.134 −0.016 −0.021
−14.295 −14.293 −13.406 −1.747 −2.356
Idiosyncratic Volatility 0.139 0.128 0.116 −0.046 −0.050
24.124 18.095 16.753 −7.138 −7.880
Skewness 0.052 0.018 −0.019 −0.037 −0.060
11.604 3.254 −3.538 −7.266 −12.018
Kurtosis −0.075 −0.047 −0.015 0.040 0.060
−15.670 −7.984 −2.523 7.513 11.314
Stock Price −0.033 −0.054 −0.055 −0.014 −0.013
−11.268 −14.903 −15.520 −1.213 −0.859
Market Beta 0.095 0.033 0.007 −0.068 −0.081
29.707 8.383 1.881 −19.061 −22.980
Log(Firm Size) −0.128 −0.104 −0.007 0.026 0.115
−31.992 −21.011 −1.348 3.415 26.088
Book-To-Market Ratio −0.058 −0.025 −0.032 0.039 0.033
−19.793 −6.940 −9.199 11.965 10.172
Past 1-Month Stock Return 0.003 0.009 0.010 0.002 0.004
0.882 2.209 2.448 0.685 1.009
Past 12-Month Stock Return 0.001 0.025 0.016 0.014 0.006
0.134 4.607 2.996 2.789 1.286
S&P500 Dummy 0.008 0.067 0.053 0.046 0.039
4.534 20.810 16.851 5.039 5.682
Monthly Turnover 0.062 −0.003 −0.005 −0.051 −0.062
23.627 −1.564 −2.537 −5.876 −8.604
Dividend Yield 0.006 0.018 0.030 0.015 0.021
0.450 3.470 3.701 2.455 2.818
Month Fixed Effects Yes Yes Yes Yes Yes
Number of Observations 113,269 113,269 113,269 113,269 113,269
Adjusted R2 0.056 0.031 0.011 0.010 0.024
This table reports the panel regression estimates for different age-based investor groups, where the excess weight
assigned to a stock in the aggregate group portfolio is the dependent variable. The group portfolio is formed by
55
combining the portfolios of all investors who belong to the group. Three aggregate group portfolios are considered:
the aggregate portfolio of younger (age range 20-38), older (age range 60-94), and very old (age range 75-94)
investors. The excess portfolio weight allocated to stock i in month t is given by: EW ipt =wipt−wimt
wimt× 100,
where, wipt is the actual weight assigned to stock i in group portfolio p in month t and wimt is the weight of stock
i in the aggregate market portfolio in month t. The mean return, idiosyncratic volatility, skewness, kurtosis,
and the price of the stock is used as independent variables. Additionally, the following stock characteristics are
employed as control variables: (i) market beta, which is estimated using past 60 months of data, (ii) firm size,
(iii) book-to-market ratio, (iv) short-term momentum (past one-month stock return), (v) longer-term momentum
(past twelve-month stock return), (vi) an S&P500 dummy which is set to one if the stock belongs to the S&P500
index, (vii) monthly volume turnover, and (viii) the annual dividend yield. The t-statistics for the coefficient
estimates are shown in smaller font below the estimates, where we use the non-parametric approach of Driscoll
and Kraay (1998) to obtain corrected standard errors. We winsorize all variables at their 0.5 and 99.5 percentile
levels and the independent variables are standardized.
56
TABLE III. – RULE OF THUMB CROSS-SECTIONAL REGRESSION ESTIMATES
Variable (1): NSTKS (2): NSTKS (3): TURN (4): TURN (5): TAX (6): PDY
Panel A: Main Rules of Thumb Regression Estimates
Intercept 4.308 4.392 5.344 4.121 0.157 1.773
18.967 20.715 27.395 11.525 14.211 18.467
Age 0.476 0.186 −0.407 −0.183 0.016 0.116
20.150 7.454 −7.347 −4.355 5.696 10.537
Investment Experience 0.517 0.423 −0.223 −0.177 0.006 −0.013
21.294 19.113 −3.922 −4.749 2.311 −2.285
Income 0.020 −0.100 0.003 −0.035
2.019 −2.665 1.329 −3.520
Education −0.035 −0.014 −0.004 −0.001
−1.878 −0.376 −1.685 −0.127
Male Dummy 0.013 0.157 0.006 −0.040
1.345 4.115 2.655 −0.969
Retired Dummy 0.005 0.006 −0.001 0.032
0.210 0.168 −0.333 3.265
Portfolio Size 1.642 0.282 0.075 0.116
30.266 7.154 22.960 11.218
Portfolio Turnover −0.399 0.093 −0.066
−17.697 28.848 −6.589
Portfolio Dividend Yield 0.152 −0.286 −0.003
5.922 −6.589 −1.221
Portfolio RMRF Exposure −0.020 0.509 0.009 −0.215
−0.821 12.291 3.149 −19.833
Portfolio SMB Exposure 0.304 0.828 0.021 −0.814
10.491 17.023 6.684 −20.737
Portfolio HML Exposure 0.052 −0.609 −0.011 0.666
1.903 −13.152 −3.695 19.019
Portfolio UMD Exposure 0.051 0.132 −0.001 −0.116
2.129 3.242 −0.071 −10.827
Mutual Fund Holdings 0.127 0.215 0.010 0.027
6.276 5.885 3.987 2.723
Number of Investors 27,716 19,906 27,716 19,906 19,906 19,906
Adjusted R2 0.044 0.252 0.035 0.158 0.120 0.282
Continued. . .
57
Dependent Variable Intercept Age Experience
Panel B: Other Rules of Thumb Regression Estimates
(1) Number of Asset Classes 2.545 0.139 0.073
24.956 13.035 6.871
(2) Mutual Fund Portfolio Expense Ratio × 100 0.856 −0.026 −0.013
11.668 −9.371 −4.601
(3) Weight in Foreign Stocks × 100 3.575 0.672 1.384
17.925 3.053 4.224
(4) Adjusted Disposition Effect × 100 1.780 −7.849 −1.467
3.527 −6.632 −2.235
(5) Local Bias × 100 29.493 −0.822 −0.434
11.590 −3.866 −2.511
(6) Margin Account Dummy 0.577 −0.023 0.007
20.744 −8.178 2.531
(7) Short-Sell Dummy 0.096 −0.017 0.015
18.239 −4.330 8.125
(8) Option Trade Dummy 0.089 −0.018 0.019
15.697 −8.735 9.701
Panel C: Logit Estimates using the Dutch DNB Data
(1) Mutual Fund Participation Dummy −5.587 0.022 0.473
−12.552 4.851 4.746
(2) All Asset Classes Participation Dummy −9.788 0.053 0.613
−7.186 3.819 2.067
This table reports the estimates from cross-sectional regressions, where a measure of investment choice reflecting
a “rule of thumb” is the dependent variable. In Panel A, we consider four different measures: (i) average number
of stocks in the portfolio (NSTKS), (ii) portfolio turnover (TURN), (iii) proportion of “losers” (stock investments
where an investor has experienced a loss) realized in the month of December (TAX), and (iv) portfolio dividend
yield (PDY). All measures are obtained for each investor using their choices during the sample period. In Panel
B, we consider the following measures: (i) number of asset classes in which the investor has at least one trade
during the sample period, (ii) the sample-period expense ratio of the investor’s mutual fund portfolio, (iii) the
proportion of equity portfolio invested in foreign stocks, (iv) local bias, measured as the proportion of equity
58
portfolio that is invested in stocks within a 100-mile radius, (v) the peer-group adjusted disposition effect (ADE)
measure, defined as the difference between an investor’s actual propensity to realize gains and the expected
propensity to realize gains (see Kumar and Lim (2008) for details), (vi) margin fund dummy that is set to one
if the investor holds a margin account, (vii) short-sell dummy that is set to one if the investor holds a short
position at least once during the sample-period, and (viii) option trade dummy that is set to one if the investor
executes an option trade at least once during the sample-period. In Panel C, the dependent variable is either
(i) the mutual fund participation dummy (set to one if the investor holds any type of mutual fund) or (ii) the
“all asset classes” participation dummy (set to one if the respondent holds stocks, bonds, and mutual funds) in
the 2005 Dutch DNB Household Survey data. The independent variables used in Panels A and B are described
in Section IV.B. In Panel C, the independent variables are age, self-reported knowledge of financial markets
(proxy for investment experience), income, education level, gender, and retirement dummy. The t-statistics for
the coefficient estimates are shown in smaller font below the estimates, where robust, clustered standard errors
are used to account for potential cross-sectional dependence within zip codes. We winsorize all variables at their
0.5 and 99.5 percentile levels and the independent variables are standardized. In Panel C, we estimate a logit
model and do not standardize the variables. We report the z-statistics for the coefficient estimates based on
robust standard errors. In Panels A and B, the individual investor data are from a large U.S. discount brokerage
house for the period from 1991 to 1996. In Panel C, we use the 2005 Dutch DNB Household Survey data.
59
TABLE IV. – INVESTMENT SKILL CROSS-SECTIONAL REGRESSION ESTIMATES
SR (1-2) Alpha (3-4) Char Adj (5-8) Diff (9)
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A: Baseline Specifications
Intercept 0.101 0.102 −0.341 −0.340 0.006 −0.001 0.031 0.001 0.006
13.173 9.568 −14.285 −13.099 0.699 −0.101 0.448 0.005 0.193
Age −0.001 −0.014 −0.042 −0.051 −0.044 −0.054 −0.063 −0.045 −0.077
−0.935 −3.355 −5.047 −4.735 −7.293 −2.945 −2.257 −4.636 −3.388
Investment Experience 0.011 0.010 0.010 0.020 0.037 0.027 0.028 0.026 0.028
11.644 9.450 1.184 2.283 6.053 3.891 2.289 3.938 3.103
Over 70 Dummy −0.007 −0.025 −0.020 −0.022 −0.014 −0.019
−1.483 −2.073 −2.163 −2.886 −2.592 −2.167
Income 0.001 0.001 0.011 0.004 −0.001 0.018
0.824 0.062 1.112 0.470 −0.108 2.013
Low Income Dummy −0.002 −0.004 −0.009 −0.011 −0.010 −0.008
−1.223 −0.275 −0.773 −1.004 −0.155 −0.218
Age × Low Income −0.012 −0.025 −0.014 −0.022 −0.023 −0.006
−2.033 −2.674 −1.996 −2.242 −2.367 −1.791
Education 0.011 0.014 0.017 0.013 0.013 0.014
1.882 2.527 2.717 2.089 2.129 1.529
Low Education Dummy −0.010 −0.012 −0.014 −0.015 −0.011 −0.025
−1.809 −1.813 −1.798 −1.401 −1.636 −1.510
Age × Low Education −0.011 −0.003 −0.005 −0.004 −0.004 −0.005
−1.789 −1.250 −2.719 −2.242 −1.843 −1.942
Male Dummy −0.001 −0.001 −0.006 −0.021 −0.011 −0.013
−0.550 −0.108 −0.945 −1.306 −1.816 −0.772
Retired Dummy 0.001 0.007 0.004 −0.010 0.002 −0.006
0.072 0.760 0.504 −0.957 0.391 −0.818
Hispanic Dummy −0.002 −0.003 0.009 −0.007 −0.018 −0.008
−0.350 −0.650 0.220 −1.072 −1.667 −0.309
African American Dummy −0.005 −0.016 −0.012 −0.003 −0.014 0.006
−1.038 −1.403 −1.133 −0.879 −1.084 0.517
Age × Hispanic −0.004 −0.034 −0.025 −0.017 −0.020 −0.036
−3.568 −3.516 −3.111 −2.876 −2.524 −2.488
Age × African American 0.002 0.010 0.009 −0.007 −0.008 −0.013
0.437 0.588 1.108 −1.035 −1.562 −1.876
(For brevity, the coefficient estimates of control variables are suppressed.)
Number of Investors 27,716 19,906 27,716 19,906 27,716 19,906 1,186,835 19,906 18,367
Adjusted R2 0.015 0.055 0.014 0.254 0.028 0.090 0.061 0.086 0.055
60
Skill Skill Difference
Variable (1): Alpha (2): Char Adj (3): Net Char Adj (4): Char Adj (5): Net Char Adj
Panel B: Specifications with Quadratic Age
Intercept −0.368 −0.016 −0.013 0.008 −0.017
−18.877 −0.656 −1.089 0.190 −1.099
Age −0.044 −0.064 −0.061 −0.079 −0.073
−4.316 −6.287 −4.521 −3.371 −3.066
Age2 −0.027 −0.018 −0.020 −0.044 −0.032
−2.922 −1.838 −2.157 −2.356 −2.624
Investment Experience 0.030 0.036 0.029 0.039 0.041
2.940 6.001 4.965 3.163 3.760
Over 70 Dummy −0.024 −0.026 −0.019 −0.017 −0.014
−2.296 −2.028 −2.614 −2.899 −1.834
(For brevity, other coefficient estimates are suppressed.)
The dependent variable is either a measure of investment skill or skill difference across the two halves of the
sample period. The following measures of investment skill are considered: (i) the monthly Sharpe ratio (SR),
(ii) the four factor alpha (Alpha), and (iii) the average characteristic-adjusted monthly portfolio return (Char
Adj). The four factor alpha is obtained by fitting a four-factor time-series model to the monthly portfolio return
series of each investor over the period the investor is active. All independent variables are defined in Section
V.C. In Column (7) of Panel A, we present panel regression estimates, where the skill measure is the monthly
characteristic-adjusted returns. In this specification, the factor exposures are excluded from the set of control
variables. In Column (8) of Panel A, the skill measure is the average net, characteristic-adjusted return. The
t-statistics for the coefficient estimates are shown in smaller font below the estimates, where robust, clustered
standard errors are used to account for potential cross-sectional dependence within zip codes. We winsorize all
variables at their 0.5 and 99.5 percentile levels and the independent variables are standardized.
61
TABLE V. – DETERMINANTS OF COGNITIVE ABILITIES
Cognitive ability measure is:
Variable Verbal Quantitative Memory Combined
Intercept −0.001 −0.005 −0.003 0.001
−0.080 −0.754 −0.571 0.118
Wealth 0.049 0.020 0.012 0.031
8.038 3.271 1.963 5.628
Income 0.047 0.053 0.002 0.041
7.818 8.482 0.372 7.230
Education 0.365 0.313 0.312 0.398
25.991 16.255 17.306 25.216
Age −0.129 −0.160 −0.239 −0.211
−13.178 −15.718 −24.088 −23.039
Retired Dummy −0.081 −0.065 −0.011 −0.056
−11.393 −10.394 −1.182 −9.953
Over 70 Dummy −0.010 −0.036 −0.021 −0.032
−1.932 −3.186 −2.902 −3.199
Age × Low Income −0.086 −0.055 −0.066 −0.084
−12.142 −7.460 −9.196 −12.794
Age × Low Education 0.001 −0.014 −0.030 −0.024
0.108 −2.500 −4.140 −3.060
Number of Individuals 22,153 21,777 21,904 22,215
Adjusted R2 0.249 0.211 0.245 0.341
The dependent variable is a measure of cognitive ability. The combined cognitive ability measure is the equal-
weighted measure of the verbal, quantitative, and memory measures. Among the independent variables, Wealth
is the total net-worth of the household including real-estate; Income is the total household income; Age is the
age of the individual, Education is a categorical variable that denotes the level of education from pre-primary to
post-tertiary; Low Income dummy is set to one for investors who are in the lowest income quintile; Low Education
dummy is set to one for investors who are in the lowest education quintile; Over 70 Dummy is set to one for
individuals with age over 70. The t-statistics for the coefficient estimates are shown in smaller font below the
estimates. We winsorize all variables at their 0.5 and 99.5 percentile levels and the independent variables have
been standardized. The household data are from the 2005 wave of the Survey of Health, Aging, and Retirement
in Europe (SHARE).
62
TABLE VI. – ROBUSTNESS TEST RESULTS
Robustness Test Age Experience Age×Hisp Age×AfrAm Age×LowInc Age×LowEdu
(1) 5-Day PTBSD −0.081 0.054 −0.018 −0.037 0.015 −0.041
−2.682 2.134 −1.704 −1.446 0.581 −1.253
(2) 10-Day PTBSD −0.153 0.082 −0.016 −0.018 −0.038 −0.071
−3.790 2.430 −1.469 −1.537 −1.882 −1.621
(3) GH1 Measure −0.023 0.036 −0.042 0.011 −0.020 −0.009
−1.912 3.469 −4.002 1.071 −1.883 −1.672
(4) GH2 Measure −0.010 0.004 −0.010 0.006 −0.005 −0.002
−2.710 2.359 −3.124 0.827 −1.434 −1.536
(5) Alternative Exp. Measure −0.054 0.033 −0.024 −0.004 −0.028 −0.003
−3.690 2.390 −2.087 −0.458 −1.836 −1.381
(6) Internet Users Sub-Sample −0.063 0.048 −0.022 −0.005 −0.025 −0.014
−3.168 2.653 −1.795 −1.161 −1.893 −1.349
(7) 1991-93 Sub Sample −0.041 0.035 0.001 0.011 −0.011 −0.010
−4.739 3.769 0.520 1.137 −2.050 −1.850
(8) 1994-96 Sub Sample −0.034 0.030 −0.021 −0.014 −0.005 −0.007
−2.064 3.672 −2.767 −1.903 −1.698 −1.740
(9) Local Four-Factor Alpha −0.039 0.026 −0.037 −0.021 −0.008 −0.009
−2.738 2.504 −3.408 −1.921 −0.757 −0.702
(10) Eight-Factor Alpha −0.047 0.022 −0.023 0.009 −0.017 −0.004
−3.745 2.314 −3.652 1.124 −2.668 −1.439
(11) Investors With Large Portf −0.053 0.018 −0.032 0.007 −0.032 0.001
−4.759 1.996 −3.268 0.768 −3.099 0.088
(12) Exclude Investors From CA −0.044 0.017 −0.029 0.016 −0.032 −0.003
−2.797 2.651 −2.732 0.532 −3.006 −1.197
This table reports the skill regression estimates using the baseline specification from Table IV, Panel A. The
dependent variable is an investment skill measure. For brevity, we only report the estimates for the variables
that are related to our main hypotheses. The estimates for the control variables are suppressed. In tests (1) and
(2), we use a trading based skill measure, which is defined as the k-day post-trade buy-sell return differential
(PTBSD(k)), k = 5, 10. In tests (3) and (4), we consider the two Graham-Harvey market timing measures
(GH1 and GH2). To compute GH1, the S&P 500 futures index is levered up or down to match the volatility
63
of the investor portfolio. GH1 is the difference between the mean return of the investor portfolio and the mean
return of the volatility-matched market portfolio. GH2 is computed by levering up or down each investor’s
portfolio to match the volatility of the S&P500 futures index. GH2 is the difference between the mean return
of the volatility-matched portfolio and the return of the S&P500 futures index. In both cases, the volatilities
are computed for the time-period in which an investor is active and the investor portfolio is levered up or down
by combining it with T-bills. In test (5), we use an alternative experience measure defined using self-reported
investment experience and knowledge measures. In test (6), we only consider a sub-sample of investors who have
used the Internet to execute trades. Tests (7) and (8) estimate the skill regression for two sub-periods, where the
mean characteristic-adjusted portfolio return is the skill measure. In test (9), we use the local four-factor alpha
as a measure of investment skill. To compute the local four-factor alpha, we consider the performance of the
local part of the investor portfolio that contains stock investments in firms located within a 100 mile radius of
investor’s location. In test (10), the skill measure is the eight-factor alpha. It is the intercept from an eight-factor
model that contains the four commonly used risk factors (RMRF, SMB, HML, and UMD), three industry factors
(Pastor and Stambaugh (2002)), and a liquidity factor (Pastor and Stambaugh (2003)). In test (11), we consider
a sub-sample of investors with large portfolios relative to income. Last, in test (12), we exclude investors who are
located in California. We use the four-factor alpha as the measure of investment skill in tests (5), (6), (11), and
(12). The t-statistic for the coefficient estimates are shown in smaller font below the estimates, where robust,
clustered standard errors are used to account for potential cross-sectional dependence within zip codes.
64
FIGURE 1. – AGE, PORTFOLIO PERFORMANCE, AND CHANGE IN PERFORMANCE
Below 32 32−36 36−40 40−44 44−48 48−52 52−56 56−64 64−72 Above 72−4
−3
−2
−1
0
1
2
3
AGE
Annual
Chara
cte
rist
ic−
Adju
sted P
erf
orm
ance L
evel
and D
iffe
renti
al
Performance Level
Performance Differential
This figure shows the risk-adjusted performance level (annualized characteristic-adjusted percentage return) andthe performance differential, conditional upon age. The performance differential is the change in the performancebetween the last three and the first three years of the sample period.
FIGURE 2. – COGNITIVE AGING, PORTFOLIO SIZE, AND PORTFOLIOPERFORMANCE
All Small (Below $10K) Medium ($10K−$50K) Large (Above $50K)−3
−2
−1
0
1
2
3
4
5
6
Portfolio Size
Annual
Ris
k−
Adju
sted P
erf
orm
ance
Low Cognitive Ability
High Cognitive AbilityHigh Minus Low
This figure shows the risk-adjusted performance (annualized characteristic-adjusted percentage return), condi-tional upon portfolio size and investors’ predicted cognitive abilities. The empirical model of cognitive abilitiesestimated in Table V is used to obtain the predicted cognitive ability measures. We regress the predicted cogni-tive ability measure on investment experience and obtain a residual cognitive ability measure that is orthogonalto experience.
65